Project

Sentiment Analysis based on Multi-source Social Network Data

Updates
0 new
0
Recommendations
0 new
0
Followers
0 new
4
Reads
0 new
35

Project log

Fu Lee Wang
added a research item
Social emotion classification draws many natural language processing researchers’ attention in recent years, since analyzing user-generated emotional documents on the Web is quite useful in recommending products, gathering public opinions, and predicting election results. However, the documents that evoke prominent social emotions are usually mixed with noisy instances, and it is also challenging to capture the textual meaning of short messages. In this work, we focus on reducing the impact of noisy instances and learning a better representation of sentences. For the former, we introduce an “emotional concentration” indicator, which is derived from emotional ratings to weight documents. For the latter, we propose a new architecture named PCNN, which utilizes two cascading convolutional layers to model the word-phrase relation and the phrase-sentence relation. This model regards continuous tokens as phrases based on an assumption that neighboring words are very likely to have internal relations, and semantic feature vectors are generated based on the phrase representation. We also present a Bayesian-based model named WMCM to learn document-level semantic features. Both PCNN and WMCM classify social emotions by capturing semantic regularities in language. Experiments on two real-world datasets indicate that the quality of learned semantic vectors and the performance of social emotion classification can be improved by our models.
Fu Lee Wang
added 4 research items
With the emergence of social media services, documents that only include a few words are becoming increasingly prevalent. More and more users post short messages to express their feelings and emotions through Twitter, Flickr, YouTube and other apps. However, the sparsity of word co-occurrence patterns in short text brings new challenges to emotion detection tasks. In this paper, we propose two supervised intensive topic models to associate latent topics with emotional labels. The first model constrains topics to relevant emotions, and then generates document-topic probability distributions. The second model establishes association among biterms and emotions by topics, and then estimates word-emotion probabilities. Experiments on short text emotion detection validate the effectiveness of the proposed models.
The population of older people increases in many developed and developing countries, so that the overall structures of the populations has been changing. However, older people are one of the most disadvantaged and vulnerable groups for digital exclusion in this technocratic society. Therefore, in this article, we aims to predict the sentiments for older people when they use the cross-platform instance messaging service such as WeChat or WhatsApp. Specifically, we adopt semi-annotation approaches to obtaining their sentimental labels from the textual data in the cross-platform instance messaging service. Furthermore, we propose a lexical-based framework for predicting the sentimental labels. The findings give us insight to develop applications for the inclusion of older people in digital world.
Cross-domain sentiment classification aims to tag sentiments for a target domain by labeled data from a source domain. Due to the difference between domains, the accuracy of a trained classifier may be very low. In this paper, we propose a boosting-based learning framework named TR-TrAdaBoost for cross-domain sentiment classification. We firstly explore the topic distribution of documents, and then combine it with the unigram TrAdaBoost. The topic distribution captures the domain information of documents, which is valuable for cross-domain sentiment classification. Experimental results indicate that TR-TrAdaBoost represents documents well and boost the performance and robustness of TrAdaBoost.
Fu Lee Wang
added 4 research items
With the extensive growth of social media services, many users express their feelings and opinions through news articles, blogs and tweets/microblogs. To discover the connections between emotions evoked in a user by varied-scale documents effectively, the paper is concerned with the problem of sentiment analysis over online news. Different from previous models which treat training documents uniformly, a weighted multi-label classification model (WMCM) is proposed by introducing the concept of " emotional concentration " to estimate the weight of training documents, in addition to tackle the issue of noisy samples for each emotion. The topic assignment is also used to distinguish different emotional senses of the same word at the semantic level. Experimental evaluations using short news headlines and long documents validate the effectiveness of the proposed WMCM for sentiment prediction.
As a concise medium to describe events, short text plays an important role to convey the opinions of users. The classification of user emotions based on short text has been a significant topic in social network analysis. Neural Network can obtain good classification performance with high generalization ability. However, conventional neural networks only use a simple back-propagation algorithm to estimate the parameters, which may introduce large instabilities when training deep neural networks by random initializations. In this paper, we apply a pre-training method to deep neural networks based on restricted Boltzmann machines, which aims to gain competitive and stable classification performance of user emotions over short text. Experimental evaluations using real-world datasets validate the effectiveness of our model on the short-text sentiment classification task.
As an important medium used to describe events, the short text is effective to convey emotions and communicate affective states. In this paper, we proposed a classification method based on probabilis-tic topic model, which greatly improve the performance of sentimental categorization methods on short text. To solve the problems of sparsity and context-dependency, we extract hidden topics behind the text and associate different words by the same topic. Evaluation on sentiment detection of short text verified the effectiveness of the proposed method.