No full-text available
To read the full-text of this research,
you can request a copy directly from the author.
Semantic Textual Similarity based on Deep Learning: Towards the Automatic Paraphrastic Detection of Monolingual and Cross-Lingual Plagiarism
Abstract
Due to the vast availability of texts and documents through the Internet (web pages, data warehouses, digital documents, scanned or transcribed, etc.) has become increasingly effortless to retrieve ideas. Unluckily, this phenomenon becomes proportional to the upsurge in cases of plagiarism. Undoubtedly, appropriating content without the permission of its author or citing its references is regarded as a case of plagiarism. Furthermore, plagiarism is becoming a profound issue in many areas, and research shows that paraphrasing, especially in the cross-lingual cases of reuse, is much harder to detect. Moreover, the recent rise in readily available multilingual content on the Web and social media has increased the problem to an unprecedented scale. The main objective of this thesis was to explore several facets of plagiarism detection by advancing the field on several facades. We offered a cascade of approaches for detecting some types of plagiarism while producing high-quality results. We particularly limit ourselves to paraphrastic, translation and plagiarism of ideas. We do not focus on literal forms of plagiarism, such as copying and pasting. Instead, we worked on the scale of the document and smaller passages like citations and sentences. We tackled in this thesis and advanced state of the art on multiple types of plagiarism detection, namely on extrinsic, intrinsic and cross-lingual plagiarism detection. For the extrinsic plagiarism detection first sub-task, which is plagiarism candidate retrieval , we developed a new document-level representation that effectively represents the global and local semantic and syntactic information. The results proved that our methods compete with the state of the art models on this sub-task. We propose a citation-based pairwise document similarity based on the best-performed word and sentence embedding regarding the second sub-task. Moreover, we decided to tackle the intrinsic plagiarism detection task by proposing a new style-embedding based on the attention mechanism. Our evaluation of this approach shows processing results on detecting style changes and breaches. Our final contribution is on the cross-lingual plagiarism analysis task, where we implement and develop a new model called Cross-Lingual Graph Transformer based Analysis (CL-GTA). We introduced Graph Transformer, a new graph representation method based on the transformer architecture that uses explicit relation encoding and provides a more efficient way to represent global graph dependencies. The experimental results show that using the Graph Transformer mechanism provided our model with state-of-the-art performance with literal and paraphrastic cases of plagiarism.
ResearchGate has not been able to resolve any citations for this publication.
The rapid growth in the digital era initiates the need to inculcate and preserve the academic originality of translated texts. Cross-lingual semantic similarity is concerned with identifying the degree of similarity of textual pairs written in two different languages and determining whether they are plagiarized. Unlike existing approaches, which exploit lexical and syntax features for mono-lingual similarity, this work proposed rich semantic features extracted from cross-language textual pairs, including topic similarity, semantic role labeling, spatial role labeling, named entities recognition, bag-of-stop words, bag-of-meanings for all terms, n-most frequent terms, n-least frequent terms, and different sets of their combinations. Knowledge-based semantic networks such as BabelNet and WordNet were used for computing semantic relatedness across different languages. This paper attempts to investigate two tasks, namely, cross-lingual semantic text similarity (CL-STS) and plagiarism detection and judgement (PD) using deep neural networks, which, to the best of our knowledge, have not been implemented before for STS and PD in cross-lingual setting, and using such combination of features. For this purpose, we proposed different neural network architectures to solve the PD task as either binary classification (plagiarism/independently written), or even deeper classification (literally translated/paraphrased/summarized/independently written). Deep neural networks were also used as regressors to predict semantic connotations for CL-STS tasks. Experimental results were performed on a large number of handmade data taken from multiple sources consisting of 71,910 Arabic-English pairs. Overall, experimental results showed that using deep neural networks with rich semantic features achieves encouraging results in comparison to the baselines. The proposed classifiers and regressors tend to show comparable performances when using different architectures of neural networks, but both the binary and multi-class classifiers outperform the regressors. Finally, the evaluation and analysis of using different sets of features reflected the supremacy of deeper semantic features on the classification results.
In this paper, we propose methods for author identification task dividing into author clustering and style breach detection. Our solution to the first problem consists of locality-sensitive hashing based clustering of real-valued vectors , which are mixtures of stylometric features and bag of n-grams. For the second problem, we propose a statistical approach based on some different tf-idf features that characterize documents. Applying the Wilcoxon Signed Rank test to these features, we determine the style breaches.
Due to the paucity of self-regulated learning studies at higher education level in Saudi Arabia, the present study examined the relationshipbetween self-regulated learning and academic achievement of Community College students at King Saud University. The sample of the study was comprised of 356 students attending a preparatory year program. The study used an SRL instrument developed by Purdie et al. (1996) and validated by Ahmad (2007) for the Arab learning context, and academic achievement was measured by students’ scores in the areas of English language skills and mathematics. Results indicatedthat the study instrument was valid and reliable for use in a Saudi university environment. Furthermore, results indicated that there is a significant and positive relationshipbetween self-regulated learningand the academic achievement of students. Similarly, the constructs of SRL (i.e., goal setting and planning, keeping records and monitoring, rehearsal and memorization, and seeking social assistance), especially goal setting and planning, were found to be significantly and positively related to achievement. Additionally, SRL and its constructs, especially goal setting and planning, were found to be significant predictors of academic achievement. The implications and suggestions forfuture research are discussed.
(1) (PDF) The Relationship Between Self-Regulated Learning and Academic Achievement for a Sample of Community College Students at King Saud University. Available from: https://www.researchgate.net/publication/335766121_The_Relationship_Between_Self-Regulated_Learning_and_Academic_Achievement_for_a_Sample_of_Community_College_Students_at_King_Saud_University?origin=mail&uploadChannel=re390&reqAcc=Nora_Schierenbeck&useStoredCopy=0 [accessed Jan 23 2020].
Background
In the string correction problem, we are to transform one string into another using a set of prescribed edit operations. In string correction using the Damerau-Levenshtein (DL) distance, the permissible edit operations are: substitution, insertion, deletion and transposition. Several algorithms for string correction using the DL distance have been proposed. The fastest and most space efficient of these algorithms is due to Lowrance and Wagner. It computes the DL distance between strings of length m and n, respectively, in O(mn) time and O(mn) space. In this paper, we focus on the development of algorithms whose asymptotic space complexity is less and whose actual runtime and energy consumption are less than those of the algorithm of Lowrance and Wagner.
Results
We develop space- and cache-efficient algorithms to compute the Damerau-Levenshtein (DL) distance between two strings as well as to find a sequence of edit operations of length equal to the DL distance. Our algorithms require O(s min{m,n}+m+n) space, where s is the size of the alphabet and m and n are, respectively, the lengths of the two strings. Previously known algorithms require O(mn) space. The space- and cache-efficient algorithms of this paper are demonstrated, experimentally, to be superior to earlier algorithms for the DL distance problem on time, space, and enery metrics using three different computational platforms.
Conclusion
Our benchmarking shows that, our algorithms are able to handle much larger sequences than earlier algorithms due to the reduction in space requirements. On a single core, we are able to compute the DL distance and an optimal edit sequence faster than known algorithms by as much as 73.1% and 63.5%, respectively. Further, we reduce energy consumption by as much as 68.5%. Multicore versions of our algorithms achieve a speedup of 23.2 on 24 cores.
Presently, information retrieval can be accomplished simply and rapidly with the use of search engines. This allows users to specify the search criteria as well as specific keywords to obtain the required results. Additionally, an index of search engines has to be updated on most recent information as it is constantly changed over time. Particularly, information retrieval results as documents are typically too extensive, which affect on accessibility of the required results for searchers. Consequently, a similarity measurement between keywords and index terms is essentially performed to facilitate searchers in accessing the required results promptly. Thus, this paper proposed the similarity measurement method between words by deploying Jaccard Coefficient. Technically, we developed a measure of similarity Jaccard with Prolog programming language to compare similarity between sets of data. Furthermore, the performance of this proposed similarity measurement method was accomplished by employing precision, recall, and F-measure. Precisely, the test results demonstrated the awareness of advantage and disadvantages of the measurement which were adapted and applied to a search for meaning by using Jaccard similarity coefficient.
The prevalent approach to sequence to sequence learning maps an input sequence to a variable length output sequence via recurrent neural networks. We introduce an architecture based entirely on convolutional neural networks. Compared to recurrent models, computations over all elements can be fully parallelized during training and optimization is easier since the number of non-linearities is fixed and independent of the input length. Our use of gated linear units eases gradient propagation and we equip each decoder layer with a separate attention module. We outperform the accuracy of the deep LSTM setup of Wu et al. (2016) on both WMT'14 English-German and WMT'14 English-French translation at an order of magnitude faster speed, both on GPU and CPU.
Given the potential X-ray radiation risk to the patient, low-dose CT has attracted a considerable interest in the medical imaging field. The current main stream low-dose CT methods include vendor-specific sinogram domain filtration and iterative reconstruction, but they need to access original raw data whose formats are not transparent to most users. Due to the difficulty of modeling the statistical characteristics in the image domain, the existing methods for directly processing reconstructed images cannot eliminate image noise very well while keeping structural details. Inspired by the idea of deep learning, here we combine the autoencoder, the deconvolution network, and shortcut connections into the residual encoder-decoder convolutional neural network (RED-CNN) for low-dose CT imaging. After patch-based training, the proposed RED-CNN achieves a competitive performance relative to the-state-of-art methods in both simulated and clinical cases. Especially, our method has been favorably evaluated in terms of noise suppression, structural preservation and lesion detection.
This paper addresses the issue of source retrieval in plagiarism detection. The task of source retrieval is retrieving all plagiarized sources of a suspicious document from a source document corpus whilst minimizing retrieval costs. The classification-based methods achieved the best performance in the current researches of source retrieval. This paper points out that it is more important to cast the problem as ranking and employ learning to rank methods to perform source retrieval. Specially, it employs RankBoost and Ranking SVM to obtain the candidate plagiarism source documents. Experimental results on the dataset of PAN@CLEF 2013 Source Retrieval show that the ranking based methods significantly outperforms the baseline methods based on classification. We argue that considering the source retrieval as a ranking problem is better than a classification problem.
We present the Wright-Fisher Indian buffet process (WF-IBP), a probabilistic model for time-dependent data assumed to have been generated by an unknown number of latent features. This model is suitable as a prior in Bayesian nonparametric feature allocation models in which the features underlying the observed data exhibit a dependency structure over time. More specifically, we establish a new framework for generating dependent Indian buffet processes, where the Poisson random field model from population genetics is used as a way of constructing dependent beta processes. Inference in the model is complex, and we describe a sophisticated Markov Chain Monte Carlo algorithm for exact posterior simulation. We apply our construction to develop a nonparametric focused topic model for collections of time-stamped text documents and test it on the full corpus of NIPS papers published from 1987 to 2015.
This paper describes the creation process and statistics of the official United Nations Parallel Corpus, the first parallel corpus composed from United Nations documents published by the original data creator. The parallel corpus presented consists of manually translated UN documents from the last 25 years (1990 to 2014) for the six official UN languages, Arabic, Chinese, English, French, Russian, and Spanish. The corpus is freely available for download under a liberal license. Apart from the pairwise aligned documents, a fully aligned subcorpus for the six official UN languages is distributed. We provide baseline BLEU scores of our Moses-based SMT systems trained with the full data of language pairs involving English and for all possible translation directions of the six-way subcorpus.
In this paper, we describe a novel approach to intrinsic plagiarism detection. Each suspicious document is divided into a series of consecutive, potentially overlapping 'windows' of equal size. These are represented by vectors containing the relative frequencies of a predetermined set of high-frequency character trigrams. Subsequently, a distance matrix is set up in which each of the document's windows is compared to each other window. The distance measure used is a symmetric adaptation of the normalized distance (nd1) proposed by Stamatatos [17]. Finally, an algorithm for outlier detection in multivariate data (based on Principal Components Analysis) is applied to the distance matrix in order to detect plagiarized sections. In the PAN-PC-2011 competition, this system (second place) achieved a competitive recall (.4279) but only reached a plagdet of .1679 due to a disappointing precision (.1075).
We describe the design and use of the Stanford CoreNLP toolkit, an extensible pipeline that provides core natural language analysis. This toolkit is quite widely used, both in the research NLP community and also among commercial and government users of open source NLP technology. We suggest that this follows from a simple, approachable design, straight-forward interfaces, the inclusion of robust and good quality analysis components, and not requiring use of a large amount of associated baggage.
This paper presents a simple unsupervised learning algorithm for recognizing synonyms, based on statistical data acquired by querying a Web search engine. The algorithm, called PMI-IR, uses Pointwise Mutual Information (PMI) and Information Retrieval (IR) to measure the similarity of pairs of words. PMI-IR is empirically evaluated using 80 synonym test questions from the Test of English as a Foreign Language (TOEFL) and 50 synonym test questions from a collection of tests for students of English as a Second Language (ESL). On both tests, the algorithm obtains a score of 74%. PMI-IR is contrasted with Latent Semantic Analysis (LSA), which achieves a score of 64% on the same 80 TOEFL questions. The paper discusses potential applications of the new unsupervised learning algorithm and some implications of the results for LSA and LSI (Latent Semantic Indexing).
Cross language plagiarism is the unacknowledged reuse of text across language pairs. It occurs if a passage of text is translated from source language to target language and no proper citation is provided. Although various methods have been developed for detection of cross language plagiarism, less attention has been paid to measure and compare their performance, especially when tackling with different types of paraphrasing through translation. In this paper, we investigate various approaches to cross language plagiarism detection. Moreover, we present a novel approach to cross language plagiarism detection using word embedding methods and explore its performance against other state-of-the-art plagiarism detection algorithms. In order to evaluate the methods, we have constructed an English-Persian bilingual plagiarism detection corpus (referred to as HAMTA-CL) comprised of seven types of obfuscation. The results show that the word embedding approach outperforms the other approaches with respect to recall when encountering heavily paraphrased passages. On the other hand, translation based approach performs well when the precision is the main consideration of the cross language plagiarism detection system.
The discovery of new materials can bring enormous societal and technological progress. In this context, exploring completely the large space of potential materials is computationally intractable. Here, we review methods for achieving inverse design, which aims to discover tailored materials from the starting point of a particular desired functionality. Recent advances from the rapidly growing field of artificial intelligence, mostly from the subfield of machine learning, have resulted in a fertile exchange of ideas, where approaches to inverse molecular design are being proposed and employed at a rapid pace. Among these, deep generative models have been applied to numerous classes of materials: rational design of prospective drugs, synthetic routes to organic compounds, and optimization of photovoltaics and redox flow batteries, as well as a variety of other solid-state materials.
Knowledge base completion (KBC) aims to predict missing information in a knowledge base. In this paper, we address the out-of-knowledge-base (OOKB) entity problem in KBC: how to answer queries concerning test entities not observed at training time. Existing embedding-based KBC models assume that all test entities are available at training time, making it unclear how to obtain embeddings for new entities without costly retraining. To solve the OOKB entity problem without retraining, we use graph neural networks (Graph-NNs) to compute the embeddings of OOKB entities, exploiting the limited auxiliary knowledge provided at test time. The experimental results show the effectiveness of our proposed model in the OOKB setting. Additionally, in the standard KBC setting in which OOKB entities are not involved, our model achieves state-of-the-art performance on the WordNet dataset.
The easy and fast access to the web and the massive existence of databases of information systems today have led to an agile increase in the phenomenon of plagiarism as a serious problem for publishers and researchers. Fact, a number of researchers have discussed this problem by adopting several techniques that can detect plagiarism. Nevertheless, most of these techniques are still insufficient for the detection of intelligent plagiarism, which still needs to be improved. In this paper, we give an overview of the best-known methods of detection of plagiarism that exist. We start with defining the concept of plagiarism and its various forms most used by plagiarists. A thorough study of these approaches is then carried out, by establishing a comparative table of these approaches according to several criteria. Moreover, we finish by defining the concept of big data as well as one of these techniques that called Text mining, which applied in the phase of extraction of documents sources for plagiarism detection.
The dominant sequence transduction models are based on complex recurrent or convolutional neural networks in an encoder-decoder configuration. The best performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.0 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data.
Over the last years, time-efficient approaches for the discovery of links between knowledge bases have been regarded as a key requirement towards implementing the idea of a Data Web. Thus, efficient and effective measures for comparing the labels of resources are central to facilitate the discovery of links between datasets on the Web of Data as well as their integration and fusion. We present a novel time-efficient implementation of filters that allow for the efficient execution of bounded Jaro- Winkler measures. We evaluate our approach on several datasets derived from DBpedia 3.9 and LinkedGeoData and containing up to 10⁶ strings and show that it scales linearly with the size of the data for large thresholds. Moreover, we also show that our approach can be easily implemented in parallel. We also evaluate our approach against SILK and show that we outperform it even on small datasets.
Continuous word representations, trained on large unlabeled corpora are useful for many natural language processing tasks. Many popular models to learn such representations ignore the morphology of words, by assigning a distinct vector to each word. This is a limitation, especially for morphologically rich languages with large vocabularies and many rare words. In this paper, we propose a new approach based on the skip-gram model, where each word is represented as a bag of character n-grams. A vector representation is associated to each character n-gram, words being represented as the sum of these representations. Our method is fast, allowing to train models on large corpus quickly. We evaluate the obtained word representations on five different languages, on word similarity and analogy tasks.
The task of (monolingual) text alignment consists in finding similar text fragments between two given documents. It has applications in plagiarism detection, detection of text reuse, author identification, authoring aid, and information retrieval, to mention only a few. We describe our approach to the text alignment subtask of the plagiarism detection competition at PAN 2014, which resulted in the best-performing system at the PAN 2014 competition and outperforms the best-performing system of the PAN 2013 competition by the cumulative evaluation measure Plagdet. Our method relies on a sentence similarity measure based on a tf-idf-like weighting scheme that permits us to consider stopwords without increasing the rate of false positives. We introduce a recursive algorithm to extend the ranges of matching sentences to maximal length passages. We also introduce a novel filtering method to resolve overlapping plagiarism cases. Our system is available as open source.
Understanding emotions is an important aspect of personal development and growth, and as such it is a key tile for the emulation of human intelligence. Besides being important for the advancement of AI, emotion processing is also important for the closely related task of polarity detection. The opportunity to automatically capture the general public's sentiments about social events, political movements, marketing campaigns, and product preferences has raised interest in both the scientific community, for the exciting open challenges, and the business world, for the remarkable fallouts in marketing and financial market prediction. This has led to the emerging fields of affective computing and sentiment analysis, which leverage human-computer interaction, information retrieval, and multimodal signal processing for distilling people's sentiments from the ever-growing amount of online social data.
Nowadays, Internet is the main source to get information from blogs, encyclopedias, discussion forums, source code repositories, and more resources which are available just one click away. The temptation to re-use these materials is very high. Even source codes are easily available through a simple search on the Web. There is a need of detecting potential instances of source code re-use. Source code re-use detection has usually been approached comparing source codes in their compiled version. When dealing with cross-language source code re-use, traditional approaches can deal only with the programming languages supported by the compiler. We assume that a source code is a piece of text,with its syntax and structure, so we aim at applying models for free text re-use detection to source code. In this paper we compare a Latent Semantic Analysis (LSA) approach with previously used text re-use detection models for measuring cross-language similarity in source code. The LSA-based approach shows slightly better results than the other models, being able to distinguish between re-used and related source codes with a high performance.
Cross-language plagiarism detection aims to detect plagiarised fragments of text among documents in different languages. In this paper, we perform a systematic examination of Cross-language Knowledge Graph Analysis; an approach that represents text fragments using knowledge graphs as a language independent content model. We analyse the contributions to cross-language plagiarism detection of the different aspects covered by knowledge graphs: word sense disambiguation, vocabulary expansion, and representation by similarities with a collection of concepts. In addition, we study both the relevance of concepts and their relations when detecting plagiarism. Finally, as a key component of the knowledge graph construction, we present a new weighting scheme of relations between concepts based on distributed representations of concepts. Experimental results in Spanish–English and German–English plagiarism detection show state-of-the-art performance and provide interesting insights on the use of knowledge graphs.
One of the key problems in spoken language understanding (SLU) is the task of slot filling. In light of the recent success of applying deep neural network technologies in domain detection and intent identification, we carried out an in-depth investigation on the use of recurrent neural networks for the more difficult task of slot filling involving sequence discrimination. In this work, we implemented and compared several important recurrent-neural-network architectures, including the Elman-type and Jordan-type recurrent networks and their variants. To make the results easy to reproduce and compare, we implemented these networks on the common Theano neural network toolkit, and evaluated them on the ATIS benchmark. We also compared our results to a conditional random fields (CRF) baseline. Our results show that on this task, both types of recurrent networks outperform the CRF baseline substantially, and a bi-directional Jordantype network that takes into account both past and future dependencies among slots works best, outperforming a CRFbased baseline by 14% in relative error reduction.
A text can be plagiarised in different ways. The text may be copied and pasted word by word, parts of the text may be changed, or the whole text may be summarised into one or two lines. Different kinds of plagiarism require different strategies to detect them. But rarely do we know beforehand what type of plagiarism we are dealing with. In this paper we present a system that can detect verbatim plagiarism and obfuscated plagiarism with similar accuracy. Our system uses three different types of n-grams: stopword n-grams, n-grams with at least one named entity, and all words n-grams. After detecting and merging the detections to obtain passages, the system finds new detections incrementally based on the idea that the passages on the vicinity of plagiarised passages are more likely to be plagiarised. The system performs well on verbatim as well as obfuscated plagiarism cases.
Deep neural nets with a large number of parameters are very powerful machine learning systems. However, overfitting is a serious problem in such networks. Large networks are also slow to use, making it difficult to deal with overfitting by combining the predictions of many different large neural nets at test time. Dropout is a technique for addressing this problem. The key idea is to randomly drop units (along with their connections) from the neural network during training. This prevents units from co-adapting too much. During training, dropout samples from an exponential number of different "thinned" networks. At test time, it is easy to approximate the effect of averaging the predictions of all these thinned networks by simply using a single unthinned network that has smaller weights. This significantly reduces overfitting and gives major improvements over other regularization methods. We show that dropout improves the performance of neural networks on supervised learning tasks in vision, speech recognition, document classification and computational biology, obtaining state-of-the-art results on many benchmark data sets. © 2014 Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever and Ruslan Salakhutdinov.
Dirichlet process (DP) mixture models are the cornerstone of non-parametric Bayesian statistics, and the development of Monte-Carlo Markov chain (MCMC) sampling methods for DP mixtures has enabled the application of non-parametric Bayesian methods to a variety of practical data analysis problems. However, MCMC sampling can be prohibitively slow,and it is important to explore alternatives.One class of alternatives is provided by variational methods, a class of deterministic algorithms that convert inference problems into optimization problems (Opper and Saad 2001; Wainwright and Jordan 2003).Thus far, variational methods have mainly been explored in the parametric setting, in particular within the formalism of the exponential family (Attias2000; Ghahramani and Beal 2001; Bleietal .2003).In this paper, we present a variational inference algorithm for DP mixtures.We present experiments that compare the algorithm to Gibbs sampling algorithms for DP mixtures of Gaussians and present an application to a large-scale image analysis problem.
The overwhelming majority of scientific publications are authored by multiple persons; yet, bibliographic metrics are only assigned to individual articles as single entities. In this paper, we aim at a more fine-grained analysis of scientific authorship. We therefore adapt a text segmentation algorithm to identify potential author changes within the main text of a scientific article, which we obtain by using existing PDF extraction techniques. To capture stylistic changes in the text, we adopt a number of stylometric features. We evaluate our approach on a small subset of PubMed articles consisting of an approximately equal number of research articles written by a varying number of authors. Our results indicate that the more authors an article has the more potential author changes are identified. These results can be considered as an initial step towards a more detailed analysis of scientific authorship, thereby extending the repertoire of bibliometrics.
Dans ce document, nous présentons certaines approches permettant de comparer des textes. Nous parlerons par la suite de similarité entre textes ou entre documents. Les approches présentées ont été sélectionnées pour répondre au mieux au contexte (détaillé dans la section suivante). Ainsi, ce document ne prétend pas donner une liste exhaustive de toutes les méthodes existantes, mais tente de donner un aperçu des méthodes les plus utilisées dans le contexte de notre étude.
In this paper, we propose a variety of Long Short-Term Memory (LSTM) based
models for sequence tagging. These models include LSTM networks, bidirectional
LSTM (BI-LSTM) networks, LSTM with a Conditional Random Field (CRF) layer
(LSTM-CRF) and bidirectional LSTM with a CRF layer (BI-LSTM-CRF). Our work is
the first to apply a bidirectional LSTM CRF (denoted as BI-LSTM-CRF) model to
NLP benchmark sequence tagging data sets. We show that the BI-LSTM-CRF model
can efficiently use both past and future input features thanks to a
bidirectional LSTM component. It can also use sentence level tag information
thanks to a CRF layer. The BI-LSTM-CRF model can produce state of the art (or
close to) accuracy on POS, chunking and NER data sets. In addition, it is
robust and has less dependence on word embedding as compared to previous
observations.
We present in this paper our system developed for SemEval 2015 Shared Task 2 (2a-En-glish Semantic Textual Similarity, STS, and 2c-Interpretable Similarity) and the results of the submitted runs. For the English STS subtask, we used regression models combining a wide array of features including semantic similarity scores obtained from various methods. One of our runs achieved weighted mean correlation score of 0.784 for sentence similarity subtask (i.e., English STS) and was ranked tenth among 74 runs submitted by 29 teams. For the interpretable similarity pilot task, we employed a rule-based approach blended with chunk alignment labeling and scoring based on semantic similarity features. Our system for interpretable text similarity was among the top three best performing systems.