Leveraging Multiple Domains for Sentiment Classification

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2978–2988, Osaka, Japan, December 11-17 2016. Leveraging Multiple Domains for Sentiment Classification Fan Yang Arjun Mukherjee Yifan Zhang Department of Computer Science, University of Houston, TX, USA fyang11@uh.edu arjun4787@gmail.com aeryen@gmail.com Abstract Sentiment classification becomes more and more important with the rapid growth of user- generated content. However, sentiment classification task usually comes with two challenges: first, sentiment classification is highly domain-dependent and training sentiment classifier for every domain is inefficient and often impractical; second, since the quantity of labeled data is important for assessing the quality of classifier, it is hard to evaluate classifiers when labeled data is limited for certain domains. To address the challenges mentioned above, we focus on learning high-level features that are able to generalize across domains, so a global classifier can benefit with a simple combination of documents from multiple domains. In this paper, the pro- posed model incorporates both labeled and unlabeled data from multiple domains and learns new feature representations. Our model doesn’t require labels from every domain, which means the learned feature representation can be generalized for sentiment domain adaptation. In addition, the learned feature representation can be used as classifier since our model defines the mean- ing of feature value and arranges high-level features in a prefixed order, so it is not necessary to train another classifier on top of the new features. Empirical evaluations demonstrate our model outperforms baselines and yields competitive results to other state-of-the-art works on the benchmark dataset. 1 Introduction With the rapid growth of user-generated content, such as product reviews and microblogs, sentiment analysis and opinion mining have become more and more important as they address the problem of analyzing user’s opinions, emotions, sentiments and attitudes. The applications of sentiment analysis have been found in almost every business and social domain (Liu, 2012; Bollen et al., 2011; Ku et al., 2006). Document-level sentiment classification predicts sentiment polarities for a given document or review. The large number of reviews not only help customers make better decisions but also make it possible yet challenge for product manufacturers to keep track opinions of the products (Hu and Liu, 2004). While machine learning techniques provide interesting methods for analyzing sentiments (Turney, 2002; Go et al., 2009; Pang et al., 2002), challenges also arise certain limitations for the development of sentiment classification. For example, sentiment expression is highly domain-dependent (Pang and Lee, 2008), but training sentiment classifier for every domain is inefficient and often impractical. Simply combining data from different domains may not contribute to a generalized classifier for every domain, as users express the same sentiment in different domains using different words, or even express different sentiment using same words. For example, a user would prefer computer or car to “run fast” but not wish to use battery “die fast” or to see watch “move fast”. A book or a movie can attract people by “unpredictable” endings but “unpredictable” economic trends scare away investors. In addition, …

Content truncated to 500 words