Article
To read the full-text of this research, you can request a copy directly from the authors.

Abstract

The use of robo-readers to analyze news texts is an emerging technology trend in computational finance. In recent research, a substantial effort has been invested to develop sophisticated financial polarity-lexicons that can be used to investigate how financial sentiments relate to future company performance. However, based on experience from other fields, where sentiment analysis is commonly applied, it is well-known that the overall semantic orientation of a sentence may differ from the prior polarity of individual words. The objective of this article is to investigate how semantic orientations can be better detected in financial and economic news by accommodating the overall phrase-structure information and domain-specific use of language. Our three main contributions are: (1) establishment of a human-annotated finance phrase-bank, which can be used as benchmark for training and evaluating alternative models; (2) presentation of a technique to enhance financial lexicons with attributes that help to identify expected direction of events that affect overall sentiment; (3) development of a linearized phrase-structure model for detecting contextual semantic orientations in financial and economic news texts. The relevance of the newly added lexicon features and the benefit of using the proposed learning-algorithm are demonstrated in a comparative study against previously used general sentiment models as well as the popular word frequency models used in recent financial studies. The proposed framework is parsimonious and avoids the explosion in feature-space caused by the use of conventional n-gram features.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

... To the best of our knowledge, there have been no studies that explore the robustness of prefix tuning that reflect real-life scenarios and compare it with finetuning to identify the more robust method. Our work corrupts the financial phrasebank dataset (Malo et al., 2014), using various text corruption methods such as keyboard errors (typos), inserting random characters, deleting random words, replacing characters with OCR alternatives and replacing words with antonyms by varying percentages in each sentence. The corrupted dataset is used with two widely used pre-trained models, BERT-base (Devlin et al., 2018) and RoBERTa-large (Liu et al., 2019), under both prefix tuning and fine-tuning, to compare their performance at different noise levels. ...
... Two financial tasks are used to evaluate the performance of prefix tuning. The first task is the sentiment analysis of the Financial Phrasebank dataset (Malo et al., 2014), which is the main dataset used to compare the performance and evaluate the robustness of both prefix tuning and fine-tuning. The second task is the sentiment analysis of the Twitter Stockmarket dataset from Kaggle, Chaudhary (2020), which is also used to evaluate the performance of prefix tuning and fine-tuning. ...
... Financial Phrasebank The Financial Phrasebank dataset (Malo et al., 2014), consists of 4840 sentences from financial news articles and the sentences were manually labelled as positive, negative or neutral by 16 annotators with backgrounds in finance and business. The annotators labelled the sentences depending on whether the information from the sentence had a positive, negative or no impact on the stock prices of the company mentioned in the sentence. ...
Preprint
The invention of transformer-based models such as BERT, GPT, and RoBERTa has enabled researchers and financial companies to finetune these powerful models and use them in different downstream tasks to achieve state-of-the-art performance. Recently, a lightweight alternative (approximately 0.1% - 3% of the original model parameters) to fine-tuning, known as prefix tuning has been introduced. This method freezes the model parameters and only updates the prefix to achieve performance comparable to full fine-tuning. Prefix tuning enables researchers and financial practitioners to achieve similar results with much fewer parameters. In this paper, we explore the robustness of prefix tuning when facing noisy data. Our experiments demonstrate that fine-tuning is more robust to noise than prefix tuning -- the latter method faces a significant decrease in performance on most corrupted data sets with increasing noise levels. Furthermore, prefix tuning has high variances in the F1 scores compared to fine-tuning in many corruption methods. We strongly advocate that caution should be carefully taken when applying the state-of-the-art prefix tuning method to noisy data.
... The Financial Phrase Bank dataset [17] contains 5000 phrases in the field of finance and economics and was first used and published in 2013. It is intended for establishing new standards for modeling techniques in a financial context. ...
... The annotators were asked to consider the sentences from only an investor's point of view, that is, whether the news could have a positive, negative, or neutral impact on the stock price. Sentences that did not appear to be relevant were considered neutral [17]. ...
... Table 1. Sentences of the Financial Phrase Bank [17]. ...
Full-text available
Article
Using sentiment information in the analysis of financial markets has attracted much attention. Natural language processing methods can be used to extract market sentiment information from texts such as news articles. The objective of this paper is to extract financial market sentiment information from news articles and use the estimated sentiment scores to predict the price direction of the stock market index Standard & Poor’s 500. To achieve the best possible performance in sentiment classification, state-of-the-art bidirectional encoder representations from transformers (BERT) models are used. The pretrained transformer networks are fine-tuned on a labeled financial text dataset and applied to news articles from known providers of financial news content to predict their sentiment scores. The generated sentiment scores for the titles of the given news articles, for the (text) content of said news articles, and for the combined title-content consideration are posited against past time series information of the stock market index. To forecast the price direction of the stock market index, the predicted sentiment scores are used in a simple strategy and as features for a random forest classifier. The results show that sentiment scores based on news content are particularly useful for stock price direction prediction.
... A text dataset with reliable labels makes it possible to successfully fine-tune BERT to a classification task and evaluate its performance with metrics such as accuracy. There are various studies that have access to manually labeled news data related to the stock market (Malo, et al., 2014;Lutz, et al., 2020;Van de Kauter, et al., 2015). It would be feasible to fine-tune BERT for sentiment classification using an expert-labeled dataset. ...
... It would be feasible to fine-tune BERT for sentiment classification using an expert-labeled dataset. Araci (2019), for instance, fine-tunes a BERT-based classifier, called FinBERT, using the stock news data from Malo et al. (2014). ...
... FinBERT (Araci, 2019) is a BERT-based sentiment classification model, fine-tuned on the Financial Phrasebank data (Malo, et al., 2014), which is a 4840-sample financial news dataset labeled by the majority vote from 16 individuals with adequate financial knowhow. The sentiment labels are either positive, neutral or negative. ...
Full-text available
Preprint
This paper studies the extent at which investor sentiment contributes to cryptocurrency return prediction. Investor sentiment is extracted from news articles, Reddit posts and Tweets using BERT-based classifiers fine-tuned on this specific text data. As this data is unlabeled, a weak supervision approach by pseudo-labeling using a zero-shot classifier is used. Contribution of sentiment is then examined using a variety of machine learning models. Each model is trained on data with and without sentiment separately. The conclusion is that sentiment leads to higher prediction accuracy and additional investment profit when the models are analyzed collectively, although this does not hold true for every single model.
... To adapt the original BERT to sentiment analysis in the financial domain, Araci (2019) was the first to propose a FinBERT model by further pretraining BERT Base on the financial subset of the Reuters TRC2 corpus. The evaluation, carried out on the Financial Phrase Bank (Malo et al., 2014) and the FiQA sentiment scoring dataset (Maia et al., 2018), demonstrated that FinBERT largely outperformed all the LSTM-based baselines and was slightly better than the original model. The second FinBERT model, introduced by Yang et al. (2020), followed two different training strategies. ...
... To compare our models in the financial domain, we selected three different datasets. The Financial PhraseBank (Malo et al., 2014) is a standard dataset for sentiment classification composed of 4,840 sentences selected from financial news and annotated for Positive, Negative, and Neutral sentiment by 16 different annotators with experience in the financial domain. The dataset comes with the original annotations: for our study, we evaluated on a subset of 2,264 instances with at least 75% of annotator agreement. ...
... Understanding numerals is of key importance for the automatic processing of financial documents. (Malo et al., 2014) 2,264 \ \ 3 81 FinTextSen (Daudert et al., 2018) 2,488 \ \ 3 476 StockSen (Xing et al., 2020) 14,457 6,218 \ 2 370 Causality Detection (Mariko et al., 2020) 13,478 \ 8,580 2 1,460 FinNum-1 subtask 1/2 (Chen et al., 2019b) 4,072 457 786 7/17 48 FinNum-2 (Chen et al., 2019a) 7,187 2,109 1,044 2 120 (2) a. $FB (110.20) is starting to show some relative strength and signs of potential B/O on the daily. b. iPhone 6 may not be as secure as Apple thought.. $AAPL ...
... This study compares machine learning methods for sentiment analysis with the n-gram feature in the case of financial news and corporate headlines. Data from research [5] with a total of 4846 titles was used. The number of neutral sentiments is 2879, the number of positive sentiments is 1363, and the number of negative sentiments is 604. ...
... Sentiment marking was carried out by a group of 16 annotators with adequate business education backgrounds. The data uses research [5] with a total of 4846 news titles. The number of neutral sentiments is 2879, the number of positive sentiments is 1363, and the number of negative sentiments is 604. ...
Full-text available
Article
Sentiment analysis is currently widely used in natural language processing or information retrieval applications. Sentiment analysis analysis can provide information related to outstanding financial news headlines and provide input to the company. Positive sentiment will also have a good impact on the development of the company, but negative sentiment will damage the company's reputation. This will affect the company's development. This study compares machine learning methods on financial news headlines with n-gram feature extraction. The purpose of this study was to obtain the best method for classifying the headline sentiment of the company's financial news. The machine learning methods compared are Multinomial Naïve Bayes, Logistic Regression, Support Vector Machine, multi-layer perceptron (MLP), Stochastic Gradient Descent, and Decision Trees. The results show that the best method is logistic regression with a percentage of f1-measure, precision, and recal of 73.94 %, 73.94 %, and 74.63 %. This shows that the n-gram and machine learning features have successfully carried out sentiment analysis.
... Furthermore, they studied the performance-related effect of pre-training on other data-sets and experimented with various finetuning strategies. 2. Malo et al., 2014 [50] created the Financial PhraseBank data-set, improved the existing lexicons on the financial domain through the addition of verbs, expressions to capture the trend and studied how the trend impacts the sentiment. Based on it, they devised a lexicon-based model for sentiment classification taking into account the structure of the sentence. ...
... Accuracy F1-Score LSTM [53] 0.7100 0.6400 LPS [50] 0.7100 0.7100 HSC [43] 0.7100 0.7600 LSTM with ELMo [53] 0.7500 0.7000 BERT cased [45] 0.7600 -ULMFit [53] 0.8300 0.7900 BERT uncased [45] 0.8400 -FinBERT [53] 0.8600 0.8400 FinBERT BaseVocab cased [45] 0.8600 -FinBERT FinVocab cased [45] 0.8600 -FinBERT BaseVocab uncased [45] 0.8700 -FinBERT FinVocab uncased [45] 0.8700 -FinBERT Base [47] 0.9100 0.8900 FinBERT Large [47] 0.9400 0.9300 Proposed model 0.9800 0.9900 ...
Full-text available
Article
With the advent of transformers having attention mechanisms, the advancements in Natural Language Processing (NLP) have been manifold. However, these models possess huge complexity and enormous computational overhead. Besides, the performance of such models relies on the feature representation strategy for encoding the input text. To address these issues, we propose a novel transformer encoder architecture with Selective Learn-Forget Network (SLFN) and contextualized word representation enhanced through Parts-of-Speech Characteristics Embedding (PSCE). The novel SLFN selectively retains significant information in the text through a gated mechanism. It enables parallel processing, captures long-range dependencies and simultaneously increases the transformer’s efficiency while processing long sequences. While the intuitive PSCE deals with polysemy, distinguishes word-inflections based on context and effectively understands the syntactic as well as semantic information in the text. The single-block architecture is extremely efficient with 96.1% reduced parameters compared to BERT. The proposed architecture yields 6.8% higher accuracy than vanilla transformer architecture and appreciable improvement over various state-of-the-art models for sentiment analysis over three data-sets from diverse domains.
... For our research we have used financial phrasebank dataset. The dataset of [18] consist about 5,000 rows from financial news in English (Table I). There are two columns named as: label and sentence. ...
... According to measured metrics, CNN model with word2vec embeddings performed better than the models LPS [18], HSC [17] reported by other papers on phrasebank dataset. However, the pre-trained language model BERT shows the state of art results [4]. ...
... In our research, we replicate the CNN-static model of Kim et al. (2014) in which the Word2Vec model freezes during the training. We use Word2Vec weight trained on the 10-K corpus of 1996-2013 (Tsai et al., 2016), and train the network with the financial sentiment analysis dataset provided by Malo et al. (2014) which consists of 4,846 sentences. The model takes each sentence as input and assigns probability to each of three classes: positive, negative, and neutral. ...
... We utilize the model structure based on the original BERT model (Devlin et al., 2018) and the fine-tuned weight of Fin-BERT (Araci, 2019) trained for financial sentiment analysis. Fin-BERT is pre-trained on the subset of Reuters TRC2 dataset which includes financial press articles and fine-tuned on the financial sentiment analysis dataset provided by Malo et al. (2014), which is identical to the dataset that we use to train the network of Word2Vec model. Similarly, the model takes each sentence as its input and assigns probability to each of three classes: positive, negative, and neutral. ...
... However, general language tasks like GLUE cannot evaluate financial domain-specific models. In English, financial language tasks such as sentiment analysis (Cortis et al., 2017;Malo, Sinha, Korhonen, Wallenius, & Takala, 2014) and causality detection (Mariko et al., 2020) have been established. ...
Article
The application of natural language processing (NLP) to financial fields is advancing with an increase in the number of available financial documents. Transformer-based models such as Bidirectional Encoder Representations from Transformers (BERT) have been successful in NLP in recent years. These cutting-edge models have been adapted to the financial domain by applying financial corpora to existing pre-trained models and by pre-training with the financial corpora from scratch. In Japanese, by contrast, financial terminology cannot be applied from a general vocabulary without further processing. In this study, we construct language models suitable for the financial domain. Furthermore, we compare methods for adapting language models to the financial domain, such as pre-training methods and vocabulary adaptation. We confirm that the adaptation of a pre-training corpus and tokenizer vocabulary based on a corpus of financial text is effective in several downstream financial tasks. No significant difference is observed between pre-training with the financial corpus and continuous pre-training from the general language model with the financial corpus. We have released our source code and pre-trained models.
... Le dataset Financial Phrasebank (FPB) [100] consiste à catégoriser des phrases pour savoir si elles ont un effet positif, négatif, ou neutre en termes d'influence sur le marché financier. Il a été annoté par des professionnels du domaine. ...
Thesis
Un nombre important de modèles probabilistes connaissent une grande perte d'intérêt pour la classification avec apprentissage supervisé depuis un certain nombre d'années, tels que le Naive Bayes ou la chaîne de Markov cachée. Ces modèles, qualifiés de génératifs, sont critiqués car leur classificateur induit doit prendre en compte la loi des observations, qui peut s'avérer très complexe à apprendre quand le nombre de features de ces derniers est élevé. C'est notamment le cas en Traitement des Langues Naturelles, où les récents algorithmes convertissent des mots en vecteurs numériques de grande taille pour atteindre de meilleures performances.Au cours de cette thèse, nous montrons que tout modèle génératif peut définir son classificateur sans prendre en compte la loi des observations. Cette proposition remet en question la catégorisation connue des modèles probabilistes et leurs classificateurs induits - en classes générative et discriminante - et ouvre la voie à un grand nombre d'applications possibles. Ainsi, la chaîne de Markov cachée peut être appliquée sans contraintes à la décomposition syntaxique de textes, ou encore le Naive Bayes à l'analyse de sentiments.Nous allons plus loin, puisque cette proposition permet de calculer le classificateur d'un modèle probabiliste génératif avec des réseaux de neurones. Par conséquent, nous « neuralisons » les modèles cités plus haut ainsi qu'un grand nombre de leurs extensions. Les modèles ainsi obtenus permettant d'atteindre des scores pertinents pour diverses tâches de Traitement des Langues Naturelles tout en étant interprétable, nécessitant peu de données d'entraînement, et étant simple à mettre en production.
... Le dataset Financial Phrasebank (FPB) [100] consiste à catégoriser des phrases pour savoir si elles ont un effet positif, négatif, ou neutre en termes d'influence sur le marché financier. Il a été annoté par des professionnels du domaine. ...
Thesis
Microservice architectures contribute to building complex distributed systems as sets of independent microservices. The decoupling and modularity of distributed microservices facilitates their independent replacement and upgradeability. Since the emergence of agile DevOps and CI/CD, there is a trend towards more frequent and rapid evolutionary changes of the running microservice-based applications in response to various evolution requirements. Applying changes to microservice architectures is performed by an evolution process of moving from the current application version to a new version. The maintenance and evolution costs of these distributed systems increase rapidly with the number of microservices.The objective of this thesis is to address the following issues: How to help engineers to build a unified and efficient version management for microservices and how to trace changes in microservice-based applications? When can microservice-based applications, especially those with long-running activities, be dynamically updated without stopping the execution of the whole system? How should the safe updating be performed to ensure service continuity and maintain system consistency?In response to these questions, this thesis proposes two main contributions. The first contribution is runtime models and an evolution graph for modelling and tracing version management of microservices. These models are built at design time and used at runtime. It helps engineers abstract architectural evolution in order to manage reconfiguration deployments, and it provides the knowledge base to be manipulated by an autonomic manager middleware in various evolution activities. The second contribution is a snapshot-based approach for dynamic software updating (DSU) of microservices. The consistent distributed snapshots of microservice-based applications are constructed to be used for specifying continuity of service, evaluating the safe update conditions and realising the update strategies. The message complexity of the DSU algorithm is not the message complexity of the distributed application, but the complexity of the consistent distributed snapshot algorithm.
... 2) Datasets: The Financial Phrase-Bank (FPB) dataset [45] and the proposed SK dataset are both used to fine-tune and test our FSA system. The FPB dataset consists of 4845 financial texts which have already been annotated by financial experts. ...
Full-text available
Preprint
As an application of Natural Language Processing (NLP) techniques, financial sentiment analysis (FSA) has become an invaluable tool for investors. Its speed and accuracy can significantly impact the returns of trading strategies.With the development of deep learning and Transformer-based pre-trained models like BERT, the accuracy of FSA has been much improved, but these time-consuming big models will also slow down the computation. To boost the processing speed of the FSA system and ensure high precision, we first propose an efficient and lightweight BERT (ELBERT) along with a novel confidence-window-based (CWB) early exit mechanism. Based on ELBERT, an innovative method to accelerate text processing on the GPU platform is developed, solving the difficult problem of making the early exit mechanism work more effectively with a large input batch size. Afterward, a fast and high-accuracy FSA system is built. Experimental results show that the proposed CWB early exit mechanism achieves significantly higher accuracy than existing early exit methods on BERT under the same computation cost. Besides, our FSA system can boost the processing speed to over 1000 texts per second with sufficient accuracy by using this acceleration method, which is nearly twice as fast as the FastBERT. Hence, this system can enable modern trading systems to quickly and accurately process financial text data.
... Investigating iPrompt in sentiment classification Finally, we study the more difficult task of prompting for sentiment classification, using four popular datasets [14][15][16]. The aim is to find a dataset-specific prompt that can describe a particular sentiment classification setting. ...
Full-text available
Preprint
Large language models (LLMs) have displayed an impressive ability to harness natural language to perform complex tasks. In this work, we explore whether we can leverage this learned ability to find and explain patterns in data. Specifically, given a pre-trained LLM and data examples, we introduce interpretable autoprompting (iPrompt), an algorithm that generates a natural-language string explaining the data. iPrompt iteratively alternates between generating explanations with an LLM and reranking them based on their performance when used as a prompt. Experiments on a wide range of datasets, from synthetic mathematics to natural-language understanding, show that iPrompt can yield meaningful insights by accurately finding groundtruth dataset descriptions. Moreover, the prompts produced by iPrompt are simultaneously human-interpretable and highly effective for generalization: on real-world sentiment classification datasets, iPrompt produces prompts that match or even improve upon human-written prompts for GPT-3. Finally, experiments with an fMRI dataset show the potential for iPrompt to aid in scientific discovery. All code for using the methods and data here is made available on Github.
... Next, we selected three datasets to reduce bias in training the models. The three datasets are Sarcasm Detection [23], E-Mail classification NLP [24] and Financial phrase-bank [25]. Before training the candidate models on these datasets, the datasets were passed through a standard preprocessing pipeline. ...
... This task is cast as a multiclass classification over a set of 173 unit types associated with numerical values found in the financial news datasets of Malo et al. (2014) and Turenne et al. (2021). This task probes the semantic representations of units associated with numerals. ...
Preprint
ive text summarization has recently become a popular approach, but data hallucination remains a serious problem, including with quantitative data. We propose a set of probing tests to evaluate the efficacy of abstract summarization models' modeling of quantitative values found in the input text. Our results show that in most cases, the encoders of recent SOTA-performing models struggle to provide embeddings that adequately represent quantitative values in the input compared to baselines, and in particular, they outperform random representations in some, but surprisingly not all, cases. Under our assumptions, this suggests that the encoder's performance contributes to the quantity hallucination problem. One model type in particular, DistilBART-CDM, was observed to underperform randomly initialized representations for several experiments, and performance versus BERT suggests that standard pretraining and fine-tuning approaches for the summarization task may play a role in underperformance for some encoders.
... Nevertheless, it may still be useful to retain the embedding model at test-time to infer coefficients for previously unseen ngrams. In this work, we study five widely-used NLP classification datasets spanning different domains including classifying the emotion/presence of hate speech in tweets [31][32][33], the sentiment of sentences from financial news [34], or the sentiment of movie reviews [35,36] (see Table 1 for an overview). Note that for all datasets, the number of unigrams is quite large, often larger than the number of training samples. ...
Full-text available
Preprint
Deep learning models have achieved impressive prediction performance but often sacrifice interpretability, a critical consideration in high-stakes domains such as healthcare or policymaking. In contrast, generalized additive models (GAMs) can maintain interpretability but often suffer from poor prediction performance due to their inability to effectively capture feature interactions. In this work, we aim to bridge this gap by using pre-trained neural language models to extract embeddings for each input before learning a linear model in the embedding space. The final model (which we call Emb-GAM) is a transparent, linear function of its input features and feature interactions. Leveraging the language model allows Emb-GAM to learn far fewer linear coefficients, model larger interactions, and generalize well to novel inputs (e.g. unseen ngrams in text). Across a variety of natural-language-processing datasets, Emb-GAM achieves strong prediction performance without sacrificing interpretability. All code is made available on Github.
... This model is a BERT model, but it was explicitly trained on financial textual data, which allowed this model to identify better sentiments in texts related to financial information. Data from Financial PhraseBank were used to train the model [60]. Another model used in the study was the FinBERT-tone model (FinBERT-tone modelio nuoroda: https://huggingface.co/yiyanghkust/finbert-tone accessed on 18 August 2022). ...
Full-text available
Article
The outbreak of war and the earlier and ongoing COVID-19 pandemic determined the need for real-time monitoring of economic activity. The economic activity of a country can be defined in different ways. Most often, the country's economic activity is characterized by various indicators such as the gross domestic product, the level of employment or unemployment of the population, the price level in the country, inflation, and other frequently used economic indicators. The most popular were the gross domestic product (GDP) and industrial production. However, such traditional tools have started to decline in modern times (as the timely knowledge of information becomes a critical factor in decision making in a rapidly changing environment) as they are published with significant delays. This work aims to use the information in the Lithuanian mass media and machine learning methods to assess whether these data can be used to assess economic activity. The aim of using these data is to determine the correlation between the usual indicators of economic activity assessment and media sentiments and to forecast traditional indicators. When evaluating consumer confidence, it is observed that the forecasting of this economic activity indicator is better based on the general index of negative sentiment (comparisons with univariate time series). In this case, the average absolute percentage error is 1.3% lower. However, if all sentiments are included in the forecasting instead of the best one, the forecasting is worse and in this case the MAPE is 5.9% higher. It is noticeable that forecasting the monthly and annual inflation rate is thus best when the overall negative sentiment is used. The MAPE of the monthly inflation rate is as much as8.5% lower, while the MAPE of the annual inflation rate is 1.5% lower.
... It is built by training the BERT model further in the finance domain, using a large corpus, and fine-tuning it for financial sentiment extraction. Unlike the TextBlob, which trains sentimental classifiers with movie reviews, the FinBERT fine-tunes sentimental classifiers with the Financial PhraseBank data set (Malo et al., 2014). Thus, FinBERT can capture the sentiment of investors more precisely toward news. ...
Full-text available
Article
This paper offers an innovative approach to capture the trend of oil futures prices based on the text‐based news. By adopting natural language processing techniques, the text features obtained from online oil news catch more hidden information, improving the forecasting accuracy of oil futures prices. We find that the textual features are complementary in improving forecasting performance, both for LightGBM and benchmark models. Besides, event studies verify the asymmetric impact of positive and negative emotional shocks on oil futures prices. The generated text‐based news features robustly reduce forecasting errors, and the reduction can be maximized by incorporating all features.
... • The financial news headlines dataset (Malo et al. 2014) was labelled with the sentiment from the perspective of a retail investor and constructed based on the human-annotated finance phrase bank. The data contained 604 negative and 1363 positive headlines. ...
Full-text available
Preprint
Parliamentary and legislative debate transcripts provide an exciting insight into elected politicians' opinions, positions, and policy preferences. They are interesting for political and social sciences as well as linguistics and natural language processing (NLP). Exiting research covers discussions within individual parliaments. In contrast, we apply advanced NLP methods to a joint and comparative analysis of six national parliaments (Bulgarian, Czech, French, Slovene, Spanish, and United Kingdom) between 2017 and 2020, whose transcripts are a part of the ParlaMint dataset collection. Using a uniform methodology, we analyze topics discussed, emotions, and sentiment. We assess if the age, gender, and political orientation of speakers can be detected from speeches. The results show some commonalities and many surprising differences among the analyzed countries.
... We consider the following evaluation datasets: subj (Pang and Lee, 2004), qc (Li and Roth, 2002), yahoo answers topics , hate speech18 ( Sentiment classification aims to identify the sentiment polarity (e.g., positive, negative) conveyed in a text. We consider the following datasets for evaluation: financial phrasebank (Malo et al., 2014), mr (Pang and Lee, 2005), and sst2 (Socher et al., 2013). ...
Full-text available
Preprint
In this work, we try to decipher the internal connection of NLP technology development in the past decades, searching for essence, which rewards us with a (potential) new learning paradigm for NLP tasks, dubbed as reStructured Pre-training (RST). In such a paradigm, the role of data will be re-emphasized, and model pre-training and fine-tuning of downstream tasks are viewed as a process of data storing and accessing. Based on that, we operationalize the simple principle that a good storage mechanism should not only have the ability to cache a large amount of data but also consider the ease of access. We achieve this by pre-training models over restructured data that consist of a variety of valuable information instead of raw data after overcoming several engineering challenges. Experimentally, RST models not only surpass strong competitors (e.g., T0) on 52/55 popular datasets from a variety of NLP tasks, but also achieve superior performance in National College Entrance Examination - English (Gaokao-English),the most authoritative examination in China. Specifically, the proposed system Qin achieves 40 points higher than the average scores made by students and 15 points higher than GPT3 with 1/16 parameters. In particular, Qin gets a high score of 138.5 (the full mark is 150) in the 2018 English exam (national paper III). We have released the Gaokao Benchmark with an online submission platform. In addition, we test our model in the 2022 College Entrance Examination English that happened a few days ago (2022.06.08), and it gets a total score of 134 (v.s. GPT3's 108).
... Financial data are mostly reported in tables,but substantial information can also be found in textual form, e.g., in company filings, analyst reports, and economic news. Such information is useful in numerous financial intelligence tasks, like stock market prediction (Chen et al., 2019;Yang et al., 2019), financial sentiment analysis (Malo et al., 2014; Source code: https://github.com/nlpaueb/finer Correspondence: eleftheriosloukas@aueb.gr 2013; Akhtar et al., 2017), economic event detection (Ein-Dor et al., 2019;Jacobs et al., 2018;Zhai and Zhang, 2019), and causality analysis (Tabari et al., 2018;Izumi and Sakaji, 2019). ...
... Despite the good results, there are applications where it could be preferable to avoid dictionaries in favor of more data driven methods, which have the advantage of higher data coverage and capability of going beyond single word sentiment expression. Malo et al. (2014) provide an example of a more sophisticated supervised corpus-based approach, in which they apply a framework modeling financial sentiment expressions by a custom data set of annotated phrases. ...
Article
In this paper, we focus our attention on leveraging the information contained in financial news to enhance the performance of a bank distress classifier. The news information should be analyzed and inserted into the predictive model in the most efficient way and this task deals with the issues related to Natural Language interpretation and to the analysis of news media. Among the different models proposed for such purpose, we investigate a deep learning approach. The methodology is based on a distributed representation of textual data obtained from a model (Doc2Vec) that maps the documents and the words contained within a text onto a reduced latent semantic space. Afterwards, a second supervised feed forward fully connected neural network is trained combining news data distributed representations with standard financial figures in input. The goal of the model is to classify the corresponding banks in distressed or tranquil state. The final aim is to comprehend both the improvement of the predictive performance of the classifier and to assess the importance of news data in the classification process. This to understand if news data really bring useful information not contained in standard financial variables.
... Aract [47] used BERT for sentiment classification and regression focusing on the financial world, FinBERT. In particular, it's training was carried out through three datasets: (i) the pre-training was carried out through TRC2financial, filtering the TRC2 corpus based on financial keywords (obtaining a dataset consisting of over 29M words and 400 k phrases); (ii) FinancialPhraseBank was used as the main sentiment analysis dataset [48] and (iii) FiQA Sentiment, a dataset created for the WWW 0 18 Conference challenge. This model, despite failing in some cases (as demonstrated in cases where it has difficulty in distinguishing phrases generally used in the business environment from positive ones), turns out to be the best compared to other state-of-the art models and provides great decision support for investors. ...
Full-text available
Article
In financial markets, sentiment analysis on natural language sentences can improve forecasting. Many investors rely on information extracted from newspapers or their feelings. Therefore, this information is expressed in their language. Sentiment analysis models classify sentences (or entire texts) with their polarity (positive, negative, or neutral) and derive a sentiment score. In this paper, we use this sentiment (polarity) score to improve the forecasting of stocks and use it as a new “view” in the Black and Litterman model. This score is related to various events (both positive and negative) that have affected some stocks. The sentences used to determine the scores are taken from articles published in Financial Times (an international financial newspaper). To improve the forecast using this average sentiment score, we use a Monte Carlo method to generate a series of possible paths for several trading hours after the article was published to discretize (or approximate) the Wiener measure, which is applied to the paths and returning an exact price as results. Finally, we use the price determined in this way to calculate a yield to be used as views in a new type of “dynamic” portfolio optimization, based on hourly prices. We compare the results by applying the views obtained, disregarding the sentiment and leaving the initial portfolio unchanged.
... In lexicon-based approaches, a domain-specific or domain adaptation lexicon such as [54][55][56][57] and Sentiwordnet [16,23,24,58] respectively is constructed based on pre-defined manual rules. In the machine learning approach [59][60][61], based on the document, sentence [62], or aspect level sentiment [63], the in-domain classifiers are trained that are used in the sentiment analysis phase. However, with some pretrained models, much more embedding methods, and significant improvement in transformer-based NLP [29,30,64] are growing in FSA. ...
Article
Real-time market prediction tool tracking public opinion in specialized newsgroups and informative market data persuades investors of financial markets. Previous works mainly used lexicon-based sentiment analysis for financial markets prediction, while recently proposed transformer-based sentiment analysis promise good results for cross-domain sentiment analysis. This work considers temporal relationships between consecutive snapshots of informative market data and mood time series for market price prediction. We calculate the sentiment mood time series via the probability distribution of news embedding generated through a BERT-based transformer language model fine-tuned for financial domain sentiment analysis. We then use a deep recurrent neural network for feature extraction followed by a dense layer for price regression. We implemented our approach as an open-source API for real-time price regression. We build a corpus of financial news related to currency pairs in foreign exchange and Cryptocurrency markets. We further augment our model with informative technical indicators and news sentiment scores aligned based on news release timestamp. Results of our experiments show significant error reduction compared to the baselines. Our Financial News and Financial Sentiment Analysis RESTFul APIs are available for public use.
... Financial data are mostly reported in tables,but substantial information can also be found in textual form, e.g., in company filings, analyst reports, and economic news. Such information is useful in numerous financial intelligence tasks, like stock market prediction (Chen et al., 2019;Yang et al., 2019), financial sentiment analysis (Malo et al., 2014; Source code: https://github.com/nlpaueb/finer Correspondence: eleftheriosloukas@aueb.gr 2013; Akhtar et al., 2017), economic event detection (Ein-Dor et al., 2019;Jacobs et al., 2018;Zhai and Zhang, 2019), and causality analysis (Tabari et al., 2018;Izumi and Sakaji, 2019). ...
Full-text available
Preprint
Publicly traded companies are required to submit periodic reports with eXtensive Business Reporting Language (XBRL) word-level tags. Manually tagging the reports is tedious and costly. We, therefore, introduce XBRL tagging as a new entity extraction task for the financial domain and release FiNER-139, a dataset of 1.1M sentences with gold XBRL tags. Unlike typical entity extraction datasets, FiNER-139 uses a much larger label set of 139 entity types. Most annotated tokens are numeric, with the correct tag per token depending mostly on context, rather than the token itself. We show that subword fragmentation of numeric expressions harms BERT's performance, allowing word-level BILSTMs to perform better. To improve BERT's performance, we propose two simple and effective solutions that replace numeric expressions with pseudo-tokens reflecting original token shapes and numeric magnitudes. We also experiment with FIN-BERT, an existing BERT model for the financial domain, and release our own BERT (SEC-BERT), pre-trained on financial filings, which performs best. Through data and error analysis, we finally identify possible limitations to inspire future work on XBRL tagging.
... fi-nancial_phrasebank (Malo et al., 2014), poem_sentiment (Sheng and Uthus, 2020), medical_questions_pairs (McCreery et al., 2020), glue-mrpc (Dolan andBrockett, 2005), gluewnli (Levesque et al., 2012), climate_fever (Diggelmann et al., 2020), glue-rte (Dagan et al., 2005;Bar-Haim et al., 2006;Giampiccolo et al., 2007;Bentivogli et al., 2009), supergluecb (de Marneffe et al., 2019, sick (Marelli et al., 2014) , hate_speech18 (de Gibert et al., 2018), ethos-national_origin (Mollas et al., 2020), ethosrace (Mollas et al., 2020), ethos-religion (Mollas et al., 2020), tweet_eval-hate (Barbieri et al., 2020), tweet_eval-stance_atheism (Barbieri et al., 2020), tweet_eval-stance_feminist (Barbieri et al., 2020, quarel (Tafjord et al., 2019a), openbookqa (Mihaylov et al., 2018), qasc (Khot et al., 2020, com-monsense_qa (Talmor et al., 2019), ai2_arc (Clark et al., 2018, codah (Chen et al., 2019), supergluecopa (Gordon et al., 2012), dream (Sun et al., 2019), quartz-with_knowledge (Tafjord et al., 2019b), quartz-no_knowledge (Tafjord et al., 2019b). The choice of datasets is made following low-resource datasets in Min et al. (2021b), with the exact same set of k-shot train data using 5 random seeds. ...
Full-text available
Preprint
Large language models (LMs) are able to in-context learn -- perform a new task via inference alone by conditioning on a few input-label pairs (demonstrations) and making predictions for new inputs. However, there has been little understanding of how the model learns and which aspects of the demonstrations contribute to end task performance. In this paper, we show that ground truth demonstrations are in fact not required -- randomly replacing labels in the demonstrations barely hurts performance, consistently over 12 different models including GPT-3. Instead, we find that other aspects of the demonstrations are the key drivers of end task performance, including the fact that they provide a few examples of (1) the label space, (2) the distribution of the input text, and (3) the overall format of the sequence. Together, our analysis provides a new way of understanding how and why in-context learning works, while opening up new questions about how much can be learned from large language models through inference alone.
... The main sentiment analysis dataset used in this paper is Financial PhraseBank 5 from (Malo et al., 2014). Financial Phrasebank consists of 4845 english articles that were categorised by sentiment class and were annotated by 16 researchers with a financial background. ...
Full-text available
Preprint
Textual data in the financial domain is becoming increasingly important as the number of financial documents rapidly grows. With the progress in natural language processing (NLP), extracting valuable information has gained popularity among researchers, deep learning has boosted the development of effective financial text mining models and made significant breakthroughs in various Natural Language Processing tasks. State-of-the-art models such as BERT (Devlin et al., 2019) model developed by Google pre-trained on a large scale of unlabeled texts from Wikipedia, has shown its effectiveness by achieving good results on general domain data. However, these models are not effective enough on finance-specific language and semantics, limiting the accuracy that financial data scientists can expect from their NLP models. In this paper, we introduce FinancialBERT, a domain-specific language representation model pre-trained on large-scale financial corpora that can enhance NLP research in the financial sector. With almost the same architecture across tasks, FinancialBERT largely outperforms BERT and other state-of-the-art models in Sentiment Analysis task when pre-trained on financial corpora. Our pre-trained model FinancialBERT is freely available at: https://huggingface.co/ahmedrachid/FinancialBERT.
... Financial news 691 (Zhang, Zhao, and LeCun 2015), (Wang et al. 2018), (Malo et al. 2014). ...
Preprint
In recent years, pretrained language models have revolutionized the NLP world, while achieving state of the art performance in various downstream tasks. However, in many cases, these models do not perform well when labeled data is scarce and the model is expected to perform in the zero or few shot setting. Recently, several works have shown that continual pretraining or performing a second phase of pretraining (inter-training) which is better aligned with the downstream task, can lead to improved results, especially in the scarce data setting. Here, we propose to leverage sentiment-carrying discourse markers to generate large-scale weakly-labeled data, which in turn can be used to adapt language models for sentiment analysis. Extensive experimental results show the value of our approach on various benchmark datasets, including the finance domain. Code, models and data are available at https://github.com/ibm/tslm-discourse-markers.
... Datasets We use three financial classification datasets, including the publicly available English FINANCIAL PHRASEBANK (Malo et al., 2014), German ONE MILLION POSTS (Schabus et al., 2017), and a new Danish FINNEWS. The FINAN-CIAL PHRASEBANK is an English sentiment analysis dataset where sentences extracted from financial news and company press releases are annotated with three labels (Positive, Negative, and Neutral). ...
... Especially in the case of finance and economic texts, annotated collections are a scarce resource and many are reserved for proprietary use only. For this reason, we use the Financial Phrase Bank dataset (Malo et al., 2014) which was also used for benchmarking the pre-trained FinBERT model for sentiment analysis (Araci, 2019). The dataset includes approximately 5000 phrases/sentences from financial news texts and company press releases. ...
... To address this issue, Loughran and McDonald proposed a financial sentiment dictionary [4]. It also became a starting point for domain-specific dictionaries such as SenticNet, SentiStrength2, and Financial Polarity Lexicon (FPL) [6][7][8]. Even though the lexicon provides a fast classification of the differentiative words, rule-based approaches only achieved around a 60% accuracy rate in the study of Taj et al. [9]. ...
Full-text available
Chapter
The sentiment analysis of news and social media posts is a growing research area with advancements in natural language processing and deep learning techniques. Although various studies addressing the extraction of the sentiment score from news and other resources for specified stocks or a stock index, still there is a lack of an analysis of the sentiment in more specialized topics such as commodity news. In this paper, several natural language processing techniques with a varying range from statistical methods to deep learning-based methods were applied on the commodity news. Firstly, the dictionary-based methods were investigated with the most common dictionaries in financial sentiment analysis such as Loughran & McDonald and Harvard dictionaries. Then, statistical models have been applied to the commodity news with count vectorizer and TF-IDF. The compression-based NCD has been also included to test on the labeled data. To improve the results of the sentiment extraction, the news data was processed by deep learning-based state-of-art models such as ULMFit, Flair, Word2Vec, XLNet, and BERT. A comprehensive analysis of all tested models was held. The final analysis indicated the performance difference between the deep learning-based and statistical models for the sentiment analysis task on the commodity news. BERT has achieved superior performance among the deep learning models for the given data.
... The BERT model with finance domain knowledge [25] was then obtained by further unsupervised tuning of this BERT-BASE model using the Reuters TRC2 finance 29-million-word dataset. Finally, we used labeled finance data from Financial PhraseBank [33], which Araci [25] used to test the performance of their model, to train the BERT model to decide the sentiment of each article in the NYT dataset. The Financial PhraseBank data contains 4,840 sentences that were annotated by 16 people with finance backgrounds to judge positive, neutral, or negative tones. ...
Article
We summarized both common and novel predictive models used for stock price prediction and combined them with technical indices, fundamental characteristics and text-based sentiment data to predict S&P stock prices. A 66.18% accuracy in S&P 500 index directional prediction and 62.09% accuracy in individual stock directional prediction was achieved by combining different machine learning models such as Random Forest and LSTM together into state-of-the-art ensemble models. The data we use contains weekly historical prices, finance reports, and text information from news items associated with 518 different common stocks issued by current and former S&P 500 large-cap companies, from January 1, 2000 to December 31, 2019. Our study's innovation includes utilizing deep language models to categorize and infer financial news item sentiment; fusing different models containing different combinations of variables and stocks to jointly make predictions; and overcoming the insufficient data problem for machine learning models in time series by using data across different stocks.
Article
The increasing interest around emotions in online texts creates the demand for financial sentiment analysis. Previous studies mainly focus on coarse-grained document-/sentence-level sentiment analysis, which ignores different sentiment polarities of various targets (e.g., company entities) in a sentence. To fill the gap, from a fine-grained target-level perspective, we propose a novel Lexicon Enhanced Collaborative Network (LECN) for targeted sentiment analysis (TSA) in financial texts. In general, the model designs a unified and collaborative framework that can capture the associations of targets and sentiment cues to enhance the overall performance of TSA. Moreover, the model dynamically incorporates sentiment lexicons to guide the sentiment classification, which cultivates the model faculty of understanding financial expressions. In addition, the model introduces a message selective-passing mechanism to adaptively control the information flow between two tasks, thereby improving the collaborative effects. To verify the effectiveness of LECN, we conduct experiments on four financial datasets, including SemEVAL2017 Task5 subset1, SemEVAL2017 Task5 subset2, FiQA 2018 Task1, and Financial PhraseBank. Results show that LECN achieves improvements over the state-of-art baseline by 1.66 p.p., 1.47 p.p., 1.94 p.p., and 1.88 p.p. in terms of F1-score. A series of further analyses also indicate that LECN has a better capacity for comprehending domain-specific expressions and can achieve the mutually beneficial effect between tasks.
Conference Paper
Data analysis models tend to work with numeric features, so you must convert the string representation to numeric before applying existing models to text data. This representation is called vector representation or vector model, and the transformation process is called vectorization. In connection with the appearance in recent years of a variety of developed text vectorization methods based on neural network methods for forming words embeddings, there is a need for a comparative analysis of vectorization approaches in order to determine promising development directions. The paper contains the results of an experimental study of various approaches to vectorization of text, as well as the results of the operation of classification algorithms with different approaches of vectorization. It is shown that the use of pre-trained text vectorization models in a number of cases provides the maximum classification accuracy, as well as the fact that, as a machine learning method among the tested, logistic regression is best suited to the task.
Preprint
Delta tuning (DET, also known as parameter-efficient tuning) is deemed as the new paradigm for using pre-trained language models (PLMs). Up to now, various DETs with distinct design elements have been proposed, achieving performance on par with fine-tuning. However, the mechanisms behind the above success are still under-explored, especially the connections among various DETs. To fathom the mystery, we hypothesize that the adaptations of different DETs could all be reparameterized as low-dimensional optimizations in a unified optimization subspace, which could be found by jointly decomposing independent solutions of different DETs. Then we explore the connections among different DETs by conducting optimization within the subspace. In experiments, we find that, for a certain DET, conducting optimization simply in the subspace could achieve comparable performance to its original space, and the found solution in the subspace could be transferred to another DET and achieve non-trivial performance. We also visualize the performance landscape of the subspace and find that there exists a substantial region where different DETs all perform well. Finally, we extend our analysis and show the strong connections between fine-tuning and DETs.
Article
In this work, we evaluated research projects of French companies using Natural Language Processing. To this end, we designed a system able to estimate the probability of obtaining a research tax credit (CIR) for a project based on its technical description. This system is designed around two modules whose outputs are concatenated and fed to a fully-connected neural network that predicts the probability of success for the project. The first module uses the FastText algorithm and a Convolutional Neural Network to extract a Text Embedding vector. The second module uses an unsupervised knowledge graph extraction method and a Graph Neural Network to extract a Graph Embedding vector. The texts used as data in this study describe the research projects of companies and are written in French. Due to their high confidentiality, no similar examples exist in the literature. This data is provided by a partner consulting firm whose work consists in helping companies raise funds for their research projects. Since the methods in the literature were not effective in extracting the knowledge graphs used in the second module for our data, we present a new Knowledge Graph extraction using an unsupervised Named Entities Recognition, NER, as our contribution.
Conference Paper
Data mining is capable of giving and providing the hidden, unknown, and interesting information in terms of knowledge in healthcare industry. It is useful to form decision support systems for the disease prediction and valid diagnosis of health issues. The concepts of data mining can be used to recommend the solutions and suggestions in medical sector for precautionary measures to control the disease origin at early stage. Today, diabetes is a most common and life taking syndrome found all over theworld. The presence of diabetes itself is a cause tomany other health issues in the form of side effects in human body. In such cases when considered, a need is to find the hidden data patterns from diabetic data to discover the knowledge so as to reduce the invisible health problems that arise in diabetic patients. Many studies have shown that AssociativeClassification concept of dataminingworks well and can derive good outcomes in terms of prediction accuracy. This research work holds the experimental results of the work carried out to predict and detect the by-diseases in diabetic patients with the application of Associative Classification, and it discusses an improved algorithmic method of Associative Classification named Associative Classification using Maximum Threshold and Super Subsets (ACMTSS) to achieve accuracy in better terms. Keywords Knowledge · By-disease · Maximum threshold · Super subsets · ACMTSS · Associative Classification
Chapter
Financial sentiment analysis allows financial institutions like banks and insurance companies to better manage the credit scoring of their customers in a better way. Financial domain uses specialized mechanisms which makes sentiment analysis difficult. In this paper, we propose a pre-trained language model which can help to solve this problem with fewer labelled data. We extend on the principles of transfer learning and transformation architecture principles and also take into consideration recent outbreak of pandemics like COVID. We apply the sentiment analysis to two different sets of data. We also take smaller training set and fine-tune the same as part of the model.KeywordsDeep learningFinancial sentiment analysisTransformer architectureBidirectional encoders
Article
Sentiment analysis is a field of study that analyses people's opinions, evaluations, feelings, ratings, sentiments, and attitudes towards entities such as products, organizations, individuals, services, topics, titles, events, and qualifications. Studies on sentiment analysis problems in social media have generally adopted intelligent classification methods. However, there are conflicting and contradictory objectives to simultaneously optimize, and active research continues into developing a more effective analysis model in terms of many metrics in order to achieve effective usage. This study considers sentiment analysis as a many-objective optimization problem for the first time. For this purpose, it first proposes a Grid-based Adaptive Many-Objective Grey Wolf Optimizer (GAM-GWO) based on the Grey Wolf Optimizer algorithm. Then, it adapts this proposed method for the sentiment analysis problem in order to obtain more successful results in terms of different metrics. The study tests the performance of the proposed approach with three different data sets. Experimental results show that GAM-GWO can achieve non-dominated and competitive results in all data set classes.
Full-text available
Preprint
This paper focuses on text data augmentation for few-shot NLP tasks. The existing data augmentation algorithms either leverage task-independent heuristic rules (e.g., Synonym Replacement) or fine-tune general-purpose pre-trained language models (e.g., GPT2) using a small training set to produce new synthetic data. Consequently, these methods have trivial task-specific knowledge and are limited to yielding low-quality synthetic data for weak baselines in simple tasks. To combat this issue, we propose the Knowledge Mixture Data Augmentation Model (KnowDA): an encoder-decoder LM pretrained on a mixture of diverse NLP tasks using Knowledge Mixture Training (KoMT). KoMT is a training procedure that reformulates input examples from various heterogeneous NLP tasks into a unified text-to-text format and employs denoising objectives in different granularity to learn to generate partial or complete samples. With the aid of KoMT, KnowDA could combine required task-specific knowledge implicitly from the learned mixture of tasks and quickly grasp the inherent synthesis law of the target task through a few given instances. To the best of our knowledge, we are the first attempt to scale the number of tasks to 100+ in multi-task co-training for data augmentation. Extensive experiments show that i) KnowDA successfully improves the performance of Albert and Deberta by a large margin on the FewGLUE benchmark, outperforming previous state-of-the-art data augmentation baselines; ii) KnowDA could also improve the model performance on the few-shot NER tasks, a held-out task type not included in KoMT.
Full-text available
Conference Paper
Models are increasing in size and complexity in the hunt for SOTA. But what if those 2% increase in performance does not make a difference in a production use case? Maybe benefits from a smaller, faster model outweigh those slight performance gains. Also, equally good performance across languages in multilingual tasks is more important than SOTA results on a single one. We present the biggest, unified, multilingual collection of sentiment analysis datasets. We use these to assess 11 models and 80 high-quality sentiment datasets (out of 342 raw datasets collected) in 27 languages and included results on the internally annotated datasets. We deeply evaluate multiple setups, including fine-tuning transformer-based models for measuring performance. We compare results in numerous dimensions addressing the imbalance in both languages coverage and dataset sizes. Finally, we present some best practices for working with such a massive collection of datasets and models for a multi-lingual perspective.
Chapter
The problem of understanding how to modify the probability of success for a stage in an R&D project is still open. Primarily in cases where it is impossible to compare a project with other competitors, the probability of passing a certain phase of the experimentation is determined by taking into account only information from within the company and not from external information.In this paper, we propose to use Natural Language Processing techniques to obtain a sentiment score for the news from the outside world. In this way, we can transform sentences expressed in natural language into a numerical value which, in addition to the internal information, allows us to better “direct” the probabilities of success in a stage.KeywordsReal optionsSentiment analysisInformation revelation
Article
BERT (Bidirectional Encoder Representations from Transformers) is one of the most popular models in Natural Language Processing (NLP) for Sentiment Analysis. The main goal is to classify sentences (or entire texts) and to obtain a score in relation to their polarity: positive, negative or neutral. Recently, a Transformer-based architecture, the fine-tuned AlBERTo [1], has been introduced to determine a sentiment score in the financial sector through a specialized corpus of sentences. In this paper, we use the sentiment (polarity) score to improve the stocks forecasting. We apply the BERT model to determine the score associated to various events (both positive and negative) that have affected some stocks in the market. The sentences used to determine the scores are newspaper articles published on MilanoFinanza. We compute both the average sentiment score and the polarity, and we use a Monte Carlo method to generate (starting from the day the article was released) a series of possible paths for the next trading days, exploiting the Bayesian inference to determine a new series of bounded drift and volatility values on the basis of the score; thus, returning an exact “directed” price as a result.
Article
Fine‐grained financial sentiment analysis on news headlines is a challenging task requiring human‐annotated datasets to achieve high performance. Limited studies have tried to address the sentiment extraction task in a setting where multiple entities are present in a news headline. In an effort to further research in this area, we make publicly available SEntFiN 1.0, a human‐annotated dataset of 10,753 news headlines with entity‐sentiment annotations, of which 2,847 headlines contain multiple entities, often with conflicting sentiments. We augment our dataset with a database of over 1,000 financial entities and their various representations in news media amounting to over 5,000 phrases. We propose a framework that enables the extraction of entity‐relevant sentiments using a feature‐based approach rather than an expression‐based approach. For sentiment extraction, we utilize 12 different learning schemes utilizing lexicon‐based and pretrained sentence representations and five classification approaches. Our experiments indicate that lexicon‐based N‐gram ensembles are above par with pretrained word embedding schemes such as GloVe. Overall, RoBERTa and finBERT (domain‐specific BERT) achieve the highest average accuracy of 94.29% and F1‐score of 93.27%. Further, using over 210,000 entity‐sentiment predictions, we validate the economic effect of sentiments on aggregate market movements over a long duration.
Article
This paper establishes a new framework for assessing multimodal statistical causality between cryptocurrency market (cryptomarket) sentiment and cryptocurrency price processes. In order to achieve this we present an efficient algorithm for multimodal statistical causality analysis based on Multiple-Output Gaussian Processes. Signals from different information sources (modalities) are jointly modelled as a Multiple-Output Gaussian Process, and then using a novel approach to statistical causality based on Gaussian Processes (GP), we study linear and non-linear causal effects between the different modalities. We demonstrate the effectiveness of our approach in a machine learning application studying the relationship between cryptocurrency spot price dynamics and sentiment time-series data specific to the crypto sector, which we conjecture influences retail investor behaviour. The investor sentiment is extracted from cryptomarket news data via methods developed in the area of statistical machine learning known as Natural Language Processing (NLP). To capture sentiment, we present a novel framework for text to time-series embedding, which we then use to construct a sentiment index from publicly available news articles. We conduct a statistical analysis of our sentiment statistical index model and compare it to alternative state-of-the-art sentiment models popular in the NLP literature. In regards to the multimodal causality, the investor sentiment is our primary modality of exploration, in addition to price and a blockchain technology-related indicator (hash rate). Analysis shows that our approach is effective in modelling causal structures of variable degree of complexity between heterogeneous data sources, and illustrates the impact that certain modelling choices for the different modalities can have on detecting causality. A solid understanding of these factors is necessary to gauge cryptocurrency adoption by retail investors and provide sentiment- and technology-based insights about the cryptocurrency market dynamics.
Full-text available
Article
Recent solutions proposed for sentence-and phrase-level sentiment analysis have reflected a variety of analytical and compu-tational paradigms that include anything from nave keyword spot-ting via machine learning to full-blown logical treatments, either in pure or hybrid forms. As all appear to succeed and fail in different aspects, it is far from evident which paradigm is the optimal one for the task. In this paper, we describe a quasi-compositional senti-ment learning and parsing framework that is well-suited for exhaus-tive, uniform, and principled sentiment classification across words, phrases, and sentences. Using a hybrid approach, we model one fundamental logically defensible compositional sentiment process directly and use supervised learning to account for more complex forms of compositionality learnt from mere flat phrase-and sentence-level sentiment annotations. The proposed framework operates on quasi-compositional sentiment polarity sequences which succinctly capture the sentiment in syntactic constituents across different struc-tural levels without any conventional n-gram features. The results obtained with the initial implementation are highly encouraging and highlight a few surprising observations pertaining to role of syntactic information and sense-level sentiment ambiguity.
Full-text available
Chapter
In addition to describing facts and events, texts often communicate information about the attitude of the writer or various participants towards material being described. The most salient clues about attitude are provided by the lexical choice of the writer but, as discussed below, the organization of the text also contributes information relevant to assessing attitude. We argue that the current work in this area that concentrates mainly on the negative or positive attitude communicated by individual terms (Edmonds and Hirst, 2002; Hatzivassiloglou and McKeown, 1997; Turney and Littman, 2002; Wiebe et al., 2001) is incomplete and often gives the wrong results when implemented directly. We then describe how the base attitudinal valence of a lexical item is modified by lexical and discourse context and propose a simple, “proof of concept” implementation for some contextual shifters.
Full-text available
Conference Paper
In this paper we investigate different approaches we developed in order to classify opinion and discover opinion sources from text, using affect, opinion and attitude lexicon. We apply these approaches on the discussion topics contained in a corpus of American Congressional speech data. We propose three approaches to classifying opinion at the speech segment level, firstly using similarity measures to the affect, opinion and attitude lexicon, secondly dependency analysis and thirdly SVM machine learning. Further, we study the impact of taking into consideration the source of opinion and the consistency in the opinion expressed, and propose three methods to classify opinion at the speaker intervention level, showing improvements over the classification of individual text segments. Finally, we propose a method to identify the party the opinion belongs to, through the identification of specific affective and non-affective lexicon used in the argumentations. We present the results obtained when evaluating the different methods we developed, together with a discussion on the issues encountered and some possible solutions. We conclude that, even at a more general level, our approach performs better than trained classifiers on specific data.
Full-text available
Article
Online social networking communities may exhibit highly complex and adaptive collective behaviors. Since emotions play such an important role in human decision making, how online networks modulate human collective mood states has become a matter of considerable interest. In spite of the increasing societal importance of online social networks, it is unknown whether assortative mixing of psychological states takes place in situations where social ties are mediated solely by online networking services in the absence of physical contact. Here, we show that the general happiness, or subjective well-being (SWB), of Twitter users, as measured from a 6-month record of their individual tweets, is indeed assortative across the Twitter social network. Our results imply that online social networks may be equally subject to the social mechanisms that cause assortative mixing in real social networks and that such assortative mixing takes place at the level of SWB. Given the increasing prevalence of online social networks, their propensity to connect users with similar levels of SWB may be an important factor in how positive and negative sentiments are maintained and spread through human society. Future research may focus on how event-specific mood states can propagate and influence user behavior in "real life."
Full-text available
Article
Adequate representation of natural language semantics requires access to vast amounts of common sense and domain-specific world knowledge. Prior work in the field was based on purely statistical techniques that did not make use of background knowledge, on limited lexicographic knowledge bases such as WordNet, or on huge manual efforts such as the CYC project. Here we propose a novel method, called Explicit Semantic Analysis (ESA), for fine-grained semantic interpretation of unrestricted natural language texts. Our method represents meaning in a high-dimensional space of concepts derived from Wikipedia, the largest encyclopedia in existence. We explicitly represent the meaning of any text in terms of Wikipedia-based concepts. We evaluate the effectiveness of our method on text categorization and on computing the degree of semantic relatedness between fragments of natural language text. Using ESA results in significant improvements over the previous state of the art in both tasks. Importantly, due to the use of natural concepts, the ESA model is easy to explain to human users.
Full-text available
Article
Extracting sentiment from text is a hard semantic problem. We develop a methodology for extracting small investor sentiment from stock message boards. The algorithm comprises dierent classifier algorithms coupled together by a voting scheme. Accuracy levels are similar to widely used Bayes classifiers, but false positives are lower and sentiment accuracy higher. Time series and cross-sectional aggregation of message information improves the quality of the resultant sentiment index, particularly in the presence of slang and ambiguity. Empirical applications evidence a relationship with stock values - aggregate tech sector sentiment is found to predict stock index levels, but not at the individual stock level. The algorithms may be used to assess the impact on investor opinion of management announcements, press releases, third-party news, and regulatory changes.
Full-text available
Article
A huge number of informal messages are posted every day in social network sites, blogs and discussion forums. Emotions seem to be frequently important in these texts for expressing friendship, showing social support or as part of online arguments. Algorithms to identify sentiment and sentiment strength are needed to help understand the role of emotion in this informal communication and also to identify inappropriate or anomalous affective utterances, potentially associated with threatening behaviour to the self or others. Nevertheless, existing sentiment detection algorithms tend to be commercially-oriented, designed to identify opinions about products rather than user behaviours. This article partly fills this gap with a new algorithm, SentiStrength, to extract sentiment strength from informal English text, using new methods to exploit the de-facto grammars and spelling styles of cyberspace. Applied to MySpace comments and with a lookup table of term sentiment strengths optimised by machine learning, SentiStrength is able to predict positive emotion with 60.6% accuracy and negative emotion with 72.8% accuracy, both based upon strength scales of 1-5. The former, but not the latter, is better than baseline and a wide range of general machine learning approaches.
Full-text available
Article
While most work in sentiment analysis in the financial domain has focused on the use of content from traditional finance news, in this work we concentrate on more subjective sources of information, blogs. We aim to automatically determine the sentiment of financial bloggers towards companies and their stocks. To do this we develop a corpus of financial blogs, annotated with polarity of sentiment with respect to a number of companies. We conduct an analysis of the annotated corpus, from which we show there is a significant level of topic shift within this collection, and also illustrate the difficulty that human annotators have when annotating certain sentiment categories. To deal with the problem of topic shift within blog articles, we propose text extraction techniques to create topic-specific sub-documents, which we use to train a sentiment classifier. We show that such approaches provide a substantial improvement over full documentclassification and that word-based approaches perform better than sentence-based or paragraph-based approaches.
Article
This paper examines whether the “soft” information contained in the text of management’s quarterly earnings press releases is incrementally informative over the company’s reported “hard” earnings news. We use Diction, a textual-analysis program, to extract various dimensions of managerial net optimism from more than 20,000 corporate earnings announcements over the period 1998 to 2006 and document that unanticipated net optimism in managers’ language affects announcement period abnormal returns and predicts post earnings announcement drift. We find that it takes longer for the market to understand the implications of soft information than those of hard information. We also find that the market response varies by firm size, turnover, media and analyst coverage, and the extent to which the standard accounting model captures the underlying economics of the firm. We also show that the second moment of soft information, the level of certainty in the text, is an important determinant of contemporaneous idiosyncratic volatility, and it predicts future idiosyncratic volatility.
Book
The Handbook of News Analytics in Finance is a landmark publication bringing together the latest models and applications of News Analytics for asset pricing, portfolio construction, trading and risk control. The content of the Hand Book is organised to provide a rapid yet comprehensive understanding of this topic. Chapter 1 sets out an overview of News Analytics (NA) with an explanation of the technology and applications. The rest of the chapters are presented in four parts. Part 1 contains an explanation of methods and models which are used to measure and quantify news sentiment. In Part 2 the relationship between news events and discovery of abnormal returns (the elusive alpha) is discussed in detail by the leading researchers and industry experts. The material in this part also covers potential application of NA to trading and fund management. Part 3 covers the use of quantified news for the purpose of monitoring, early diagnostics and risk control. Part 4 is entirely industry focused; it contains insights of experts from leading technology (content) vendors. It also contains a discussion of technologies and finally a compact directory of content vendor and financial analytics companies in the marketplace of NA. The book draws equally upon the expertise of academics and practitioners who have developed these models and is supported by two major content vendors - RavenPack and Thomson Reuters - leading providers of news analytics software and machine readable news. The book will appeal to decision makers in the banking, finance and insurance services industry. In particular: asset managers; quantitative fund managers; hedge fund managers; algorithmic traders; proprietary (program) trading desks; sell-side firms; brokerage houses; risk managers and research departments will benefit from the unique insights into this new and pertinent area of financial modelling.
Article
Earnings press releases are the primary mechanism by which managers announce quarterly earnings and make other concurrent disclosures to investors and other stakeholders. A largely unexplored element of earnings press releases is the language that managers use throughout the press release, which we argue provides a unifying framework for these disclosures and an opportunity for managers to signal, both directly and more subtly, their expectations about future performance. We analyze the full texts of approximately 23,000 earnings press releases issued between 1998 and 2003 and examine whether the language used in these earnings press releases provides a signal about expected future firm performance and whether the market responds to this signal. Using categories derived from linguistic theory, we count words characterized as optimistic and pessimistic and construct a measure of managers' net optimistic language for each earnings press release. We find that this measure is positively associated with future return on assets and generates a significant market response in a short window around the earnings announcement date. We include in our models the earnings surprise as well as other quantifiable, concurrent disclosures identified in prior research as associated with the market's reaction to earnings press releases. Our results support the premise that earnings press release language provides a signal regarding managers' future earnings expectations to the market and that the market responds to this signal. We interpret our evidence to suggest that managers use language in earnings press releases to communicate credible information about expected future firm performance.
Article
The abstract for this document is available on CSA Illumina.To view the Abstract, click the Abstract button above the document title.
Article
A review of news analytics and its applications in finance is given in this chapter. In particular we review the multiple facets of current research and some of the major applications. It is widely recognised news plays a key role in financial markets. The sources and volumes of news continue to grow. New technologies that enable automatic or semi-automatic news collection, extraction, aggregation and categorisation are emerging. Further machine learning techniques can be used to process the textual input of news stories to determine quantitative sentiment scores. We consider the various types of news available and how these can be processed to form inputs to financial models. We report applications of news, for prediction of abnormal returns, for trading strategies, for diagnostic applications as well as the use of news for risk control.
Article
Most of the existing information retrieval systems are based on bag-of-words model and are not equipped with common world knowledge. Work has been done towards improving the efficiency of such systems by using intelligent algorithms to generate search queries, however, not much research has been done in the direction of incorporating human-and-society level knowledge in the queries. This paper is one of the first attempts where such information is incorporated into the search queries using Wikipedia semantics. The paper presents Wikipedia-based Evolutionary Semantics (Wiki-ES) framework for generating concept based queries using a set of relevance statements provided by the user. The query learning is handled by a co-evolving genetic programming procedure. To evaluate the proposed framework, the system is compared to a bag-of-words based genetic programming framework as well as to a number of alternative document filtering techniques. The results obtained using Reuters newswire documents are encouraging. In particular, the injection of Wikipedia semantics into a GP-algorithm leads to improvement in average recall and precision, when compared to a similar system without human knowledge. A further comparison against other document filtering frameworks suggests that the proposed GP-method also performs well when compared with systems that do not rely on query-expression learning.
Article
An increasing number of studies in political communication focus on the “sentiment” or “tone” of news content, political speeches, or advertisements. This growing interest in measuring sentiment coincides with a dramatic increase in the volume of digitized information. Computer automation has a great deal of potential in this new media environment. The objective here is to outline and validate a new automated measurement instrument for sentiment analysis in political texts. Our instrument uses a dictionary-based approach consisting of a simple word count of the frequency of keywords in a text from a predefined dictionary. The design of the freely available Lexicoder Sentiment Dictionary (LSD) is discussed in detail here. The dictionary is tested against a body of human-coded news content, and the resulting codes are also compared to results from nine existing content-analytic dictionaries. Analyses suggest that the LSD produces results that are more systematically related to human coding than are results based on the other available dictionaries. The LSD is thus a useful starting point for a revived discussion about dictionary construction and validation in sentiment analysis for political communication.
Article
We present preliminary work on classifying blog text ac-cording to the mood reported by its author during the writ-ing. Our data consists of a large collection of blog posts – online diary entries – which include an indication of the writer's mood. We obtain modest, but consistent improve-ments over a baseline; our results show that further increas-ing the amount of available training data will lead to an additional increase in accuracy. Additionally, we show that the classification accuracy, although low, is not substantially worse than human performance on the same task. Our main finding is that mood classification is a challenging task using current text analysis methods.
Article
Earnings press releases are the primary mechanism by which managers announce quarterly earnings and make other concurrent disclosures to investors and other stakeholders. A largely unexplored element of earnings press releases is the language that managers use throughout the press release, which we argue provides a unifying framework for these disclosures and an opportunity for managers to signal, both directly and more subtly, their expectations about future performance. We analyze the full texts of approximately 23,000 earnings press releases issued between 1998 and 2003 and examine whether the language used in these earnings press releases provides a signal about expected future firm performance and whether the market responds to this signal. Using categories derived from linguistic theory, we count words characterized as optimistic and pessimistic and construct a measure of managers’ net optimistic language for each earnings press release. We find that this measure is positively associated with future ROA and generates a significant market response in a short window around the earnings announcement date. We include in our models the earnings surprise as well as other quantifiable, concurrent disclosures identified in prior research as associated with the market’s reaction to earnings press releases. Our results support the premise that earnings press release language provides a signal regarding managers’ future earnings expectations to the market and that the market responds to this signal. We interpret our evidence to suggest that managers use language in earnings press releases to communicate credible information about expected future firm performance.
Article
This paper examines the tone and content of the forward-looking statements (FLS) in corporate 10-K and 10-Q filings using a Naive Bayesian machine learning algorithm. I first manually categorize 30,000 sentences of randomly selected FLS extracted from the MD&As along two dimensions: (1) tone (i.e., positive versus negative tone); and (2) content (i.e., profitability, operations, and liquidity etc.). These manually coded sentences are then used as training data in a Naive Bayesian machine learning algorithm to classify the tone and content of about 13 million forward-looking statements from more than 140,000 corporate 10-K and 10-Q MD&As between 1994 and 2007. I find that firms with better current performance, lower accruals, smaller size, lower market-to-book ratio, and less return volatility tend to have more positive forward-looking statements in MD&As. The average tone of the forward-looking statements in a firm's MD&A is positively associated with future earnings and liquidity, even after controlling for other determinants of future performance and there is no systematic change in the information content of MD&As over time. Finally, the evidence indicates that financial analysts do not fully understand the information content of the MD&As in making their forecasts.
Article
I examine the role of information processing costs on post earnings announcement drift. I distinguish between hard information - quantitative information that is more easily processed - and soft information which has higher processing costs. I find that qualitative earnings information has additional predictability for asset prices beyond the predictability in quantitative information. I also find that qualitative information has greater predictability for returns at longer horizons, suggesting that frictions in information processing generate price drift. Using a tool from natural language processing called typed dependency parsing, I demonstrate that qualitative information relating to positive fundamentals and future performance is the most difficult information to process.
Article
This paper examines the information content of the forward-looking statements (FLS) in the Management Discussion and Analysis section (MD&A) of 10-K and 10-Q filings using a Naïve Bayesian machine learning algorithm. I find that firms with better current performance, lower accruals, smaller size, lower market-to-book ratio, less return volatility, lower MD&A Fog index, and longer history tend to have more positive FLSs. The average tone of the FLS is positively associated with future earnings even after controlling for other determinants of future performance. The results also show that, despite increased regulations aimed at strengthening MD&A disclosures, there is no systematic change in the information content of MD&As over time. In addition, the tone in MD&As seems to mitigate the mispricing of accruals. When managers “warn” about the future performance implications of accruals (i.e., the MD&A tone is positive (negative) when accruals are negative (positive)), accruals are not associated with future returns. The tone measures based on three commonly used dictionaries (Diction, General Inquirer, and the Linguistic Inquiry and Word Count) do not positively predict future performance. This result suggests that these dictionaries might not work well for analyzing corporate filings.
Article
Previous research uses negative word counts to measure the tone of a text. We show that word lists developed for other disciplines misclassify common words in financial text. In a large sample of 10-Ks during 1994 to 2008, almost three-fourths of the words identified as negative by the widely used Harvard Dictionary are words typically not considered negative in financial contexts. We develop an alternative negative word list, along with five other word lists, that better reflect tone in financial text. We link the word lists to 10-K filing returns, trading volume, return volatility, fraud, material weakness, and unexpected earnings.
Article
Multifactor models are often used as a tool to describe equity portfolio risk. Naturally, risk is dependent on the market environment and investor sentiment. Traditional factor models fail to update quickly as market conditions change. It is desirable that the risk model updates to incorporate new information as it becomes available and for this reason diBartolomeo & Warrick introduce a factor model that uses option implied volatility to improve estimates of the future covariance matrix. We extend this work to use both quantified news and implied volatility to improve risk estimates as the market sentiment and environment changes.
Article
Methanol synthesis from carbon dioxide hydrogenation was studied over ceria/-alumina- and yttria-doped ceria (YDC)/-alumina-supported copper oxide catalysts to seek insight into the catalysis at metal–support interfaces. It was found that, in comparison with Cu/-Al2O3, the Cu/CeO2/-Al2O3 and Cu/YDC/-Al2O3 catalysts exhibited substantial enhancement in activity and selectivity toward methanol formation. The extent of enhancement was augmented by increased ceria loading on -alumina and with increased yttria doping into ceria. The enhancement is inferred to result from the synergistic effect between copper oxide and surface oxygen vacancies of ceria.
Article
Online communications at web portals represents technology-mediated user interactions, leading to massive data and potentially new techno-social phenomena not seen in real social mixing. Apart from being dynamically driven, the user interactions via posts is indirect, suggesting the importance of the contents of the posted material. We present a systematic way to study Blog data by combined approaches of physics of complex networks and computer science methods of text analysis. We are mapping the Blog data onto a bipartite network where users and posts with comments are two natural partitions. With the machine learning methods we classify the texts of posts and comments for their emotional contents as positive or negative, or otherwise objective (neutral). Using the spectral methods of weighted bipartite graphs, we identify topological communities featuring the users clustered around certain popular posts, and underly the role of emotional contents in the emergence and evolution of these communities.
Conference Paper
The personal, diary-like nature of blogs prompts many blog- gers to indicate their mood at the time of posting. Aggregat- ing these indications over a large amount of bloggers gives a "blogosphere state-of-mind" for each point in time: the inten- sity of different moods among bloggers at that time. In this paper, we address the task of estimating this state-of-mind from the text written by bloggers. To this end, we build mod- els that predict the levels of various moods according to the language used by bloggers at a given time; our models show high correlation with the moods actually measured, and sub- stantially outperform a baseline.
Article
Sentiment analysis is concerned with the automatic extraction of sentiment-related information from text. Although most sentiment analysis addresses commercial tasks, such as extracting opinions from product reviews, there is increasing interest in the affective dimension of the social web, and Twitter in particular. Most sentiment analysis algorithms are not ideally suited to this task because they exploit indirect indicators of sentiment that can reflect genre or topic instead. Hence, such algorithms used to process social web texts can identify spurious sentiment patterns caused by topics rather than affective phenomena. This article assesses an improved version of the algorithm SentiStrength for sentiment strength detection across the social web that primarily uses direct indications of sentiment. The results from six diverse social web data sets (MySpace, Twitter, YouTube, Digg, RunnersWorld, BBCForums) indicate that SentiStrength 2 is successful in the sense of performing better than a baseline approach for all data sets in both supervised and unsupervised cases. SentiStrength is not always better than machine-learning approaches that exploit indirect indicators of sentiment, however, and is particularly weaker for positive sentiment in news-related discussions. Overall, the results suggest that, even unsupervised, SentiStrength is robust enough to be applied to a wide variety of different social web contexts.
Article
In this article, we propose a new concept-based method for document classification. The conceptual knowledge associated with the words is drawn from Wikipedia. The purpose is to utilize the abundant semantic relatedness information available in Wikipedia in an efficient value function-based query learning algorithm. The procedure learns the value function by solving a simple linear programming problem formulated using the training documents. The learning involves a step-wise iterative process that helps in generating a value function with an appropriate set of concepts (dimensions) chosen from a collection of concepts. Once the value function is formulated, it is utilized to make a decision between relevance and irrelevance. The value assigned to a particular document from the value function can be further used to rank the documents according to their relevance. Reuters newswire documents have been used to evaluate the efficacy of the procedure. An extensive comparison with other frameworks has been performed. The results are promising.
Article
ity of the phrase in which a particular instance of a word appears may be quite different from the word's prior polarity. Positive words are used in phrases expressing negative sentiments, or vice versa. Also, quite often words that are positive or negative out of context are neutral in context, meaning they are not even being used to express a sentiment. The goal of this work is to automatically distinguish between prior and contextual polarity, with a focus on understanding which features are important for this task. Because an important aspect of the problem is identifying when polar terms are being used in neutral contexts, features for distinguishing between neutral and polar instances are evaluated, as well as features for distinguishing between positive and negative contextual polarity. The evaluation includes assessing the performance of features across multiple machine learning algorithms. For all learning algorithms except one, the combination of all features together gives the best performance. Another facet of the evaluation considers how the presence of neutral instances affects the performance of features for distinguishing between positive and negative polarity. These experiments show that the presence of neutral instances greatly degrades the performance of these features, and that perhaps the best way to improve performance across all polarity classes is to improve the system's ability to identify when an instance is neutral.
Article
This paper describes a corpus annotation project to study issues in the manual annotation of opinions, emotions, sentiments, speculations, evaluations and other private states in language. The resulting corpus annotation scheme is described, as well as examples of its use. In addition, the manual annotation process and the results of an inter-annotator agreement study on a 10,000-sentence corpus of articles drawn from the world press are presented.
Article
This paper presents a bootstrapping process that learns linguistically rich extraction patterns for subjective (opinionated) expressions. High-precision classifiers label unannotated data to automatically create a large training set, which is then given to an extraction pattern learning algorithm. The learned patterns are then used to identify more subjective sentences. The bootstrapping process learns many subjective patterns and increases recall while maintaining high precision.
Article
Noam Chomsky's first book on syntactic structures is one of the first serious attempts on the part of a linguist to construct within the tradition of scientific theory-construction a comprehensive theory of language which may be understood in the same sense that a chemical, biological theory is understood by experts in those fields. It is not a mere reorganization of the data into a new kind of library catalogue, nor another specualtive philosophy about the nature of man and language, but rather a rigorus explication of our intuitions about our language in terms of an overt axiom system, the theorems derivable from it, explicit results which may be compared with new data and other intuitions, all based plainly on an overt theory of the internal structure of languages; and it may well provide an opportunity for the application of explicity measures of simplicity to decide preference of one form over another form of grammar.
Article
This paper examines whether the "soft" information contained in the text of management's quarterly earnings press releases is incrementally informative over the company's reported "hard" earnings news. We use Diction, a textual-analysis program, to extract various dimensions of managerial net optimism from more than 20,000 corporate earnings announcements over the period 1998 to 2006 and document that unanticipated net optimism in managers' language affects announcement period abnormal returns and predicts post-earnings announcement drift. We find that it takes longer for the market to understand the implications of soft information than those of hard information. We also find that the market response varies by firm size, turnover, media and analyst coverage, and the extent to which the standard accounting model captures the underlying economics of the firm. We also show that the second moment of soft information, the level of certainty in the text, is an important determinant of contemporaneous idiosyncratic volatility, and it predicts future idiosyncratic volatility.
Article
Reliability coefficients often take the form of intraclass correlation coefficients. In this article, guidelines are given for choosing among 6 different forms of the intraclass correlation for reliability studies in which n targets are rated by k judges. Relevant to the choice of the coefficient are the appropriate statistical model for the reliability study and the applications to be made of the reliability results. Confidence intervals for each of the forms are reviewed. (23 ref) (PsycINFO Database Record (c) 2006 APA, all rights reserved).
Article
Mitochondria possess an inner membrane channel, the permeability transition pore, which is inhibited by cyclosporin A (CsA) and by matrix protons. As suggested recently by our laboratory, pore closure by these inhibitors may be due to dissociation of mitochondrial cyclophilin (CyP-M), a matrix peptidyl-prolyl-cis-trans isomerase, from its putative binding site on the pore. Unbinding of CyP-M would follow a CsA-dependent or proton-dependent change in conformation of the CyP-M molecule. It is interesting that upon binding of CsA the enzymatic activity of CyP-M is inhibited, but it is not clear whether this event plays a role in pore inhibition. Here we report experiments designed to further test the role of CyP-M in pore function. Our results indicate that CyP-M-dependent and independent mechanisms of pore activation may exist, and that the peptidylprolyl-cis-trans-isomerase activity of CyP-M is not necessarily involved in pore modulation by CyP-M.
Article
Support vector machines (SVMs) were originally designed for binary classification. How to effectively extend it for multiclass classification is still an ongoing research issue. Several methods have been proposed where typically we construct a multiclass classifier by combining several binary classifiers. Some authors also proposed methods that consider all classes at once. As it is computationally more expensive to solve multiclass problems, comparisons of these methods using large-scale problems have not been seriously conducted. Especially for methods solving multiclass SVM in one step, a much larger optimization problem is required so up to now experiments are limited to small data sets. In this paper we give decomposition implementations for two such "all-together" methods. We then compare their performance with three methods based on binary classifications: "one-against-all," "one-against-one," and directed acyclic graph SVM (DAGSVM). Our experiments indicate that the "one-against-one" and DAG methods are more suitable for practical use than the other methods. Results also show that for large problems methods by considering all data at once in general need fewer support vectors.
Article
We test and confirm the hypothesis that individual investors are net buyers of attention-grabbing stocks, e.g., stocks in the news, stocks experiencing high abnormal trading volume, and stocks with extreme one-day returns. Attention-driven buying results from the difficulty that investors have searching the thousands of stocks they can potentially buy. Individual investors do not face the same search problem when selling because they tend to sell only stocks they already own. We hypothesize that many investors consider purchasing only stocks that have first caught their attention. Thus, preferences determine choices after attention has determined the choice set.
Article
We examine whether a simple quantitative measure of language can be used to predict individual firms' accounting earnings and stock returns. Our three main findings are: (1) the fraction of negative words in firm-specific news stories forecasts low firm earnings; (2) firms' stock prices briefly underreact to the information embedded in negative words; and (3) the earnings and return predictability from negative words is largest for the stories that focus on fundamentals. Together these findings suggest that linguistic media content captures otherwise hard-to-quantify aspects of firms' fundamentals, which investors quickly incorporate into stock prices. Copyright (c) 2008 by The American Finance Association.
Article
I quantitatively measure the interactions between the media and the stock market using daily content from a popular "Wall Street Journal" column. I find that high media pessimism predicts downward pressure on market prices followed by a reversion to fundamentals, and unusually high or low pessimism predicts high market trading volume. These and similar results are consistent with theoretical models of noise and liquidity traders, and are inconsistent with theories of media content as a proxy for new information about fundamental asset values, as a proxy for market volatility, or as a sideshow with no relationship to asset markets. Copyright 2007 by The American Finance Association.