Sentiment Analysis using Support Vector Machines with Diverse Information Sources

Sentiment analysis using support vector machines with diverse information sources Tony Mullen and Nigel Collier National Institute of Informatics (NII) Hitotsubashi 2-1-2, Chiyoda-ku Tokyo 101-8430 Japan � mullen,collier � @nii.ac.jp Abstract This paper introduces an approach to sentiment analysis which uses support vector machines (SVMs) to bring together diverse sources of po- tentially pertinent information, including several fa- vorability measures for phrases and adjectives and, where available, knowledge of the topic of the text. Models using the features introduced are fur- ther combined with unigram models which have been shown to be effective in the past (Pang et al., 2002) and lemmatized versions of the unigram models. Experiments on movie review data from Epinions.com demonstrate that hybrid SVMs which combine unigram-style feature-based SVMs with those based on real-valued favorability measures obtain superior performance, producing the best re- sults yet published using this data. Further experi- ments using a feature set enriched with topic infor- mation on a smaller dataset of music reviews hand- annotated for topic are also reported, the results of which suggest that incorporating topic information into such models may also yield improvement. 1 Introduction Recently an increasing amount of research has been devoted to investigating methods of recognizing fa- vorable and unfavorable sentiments towards specific subjects within natural language texts. Areas of ap- plication for such analysis are numerous and varied, ranging from newsgroup flame filtering and infor- mative augmentation of search engine responses to analysis of public opinion trends and customer feed- back. For many of these tasks, classifying the tone of the communication as generally positive or nega- tive is an important step. There are a number of challenging aspects of this task. Opinions in natural language are very of- ten expressed in subtle and complex ways, present- ing challenges which may not be easily addressed by simple text categorization approaches such as n-gram or keyword identification approaches. Al- though such approaches have been employed effec- tively (Pang et al., 2002), there appears to remain considerable room for improvement. Moving be- yond these approaches can involve addressing the task at several levels. Recognizing the semantic im- pact of words or phrases is a challenging task in it- self, but in many cases the overarching sentiment of a text is not the same as that of decontextualized snippets. Negative reviews may contain many ap- parently positive phrases even while maintaining a strongly negative tone, and the opposite is also com- mon. This paper introduces an approach to classify- ing texts as positive or negative using Support Vec- tor Machines (SVMs), a well-known and powerful tool for classification of vectors of real-valued fea- tures (Vapnik, 1998). The present approach em- phasizes the use of a variety of diverse information sources, and SVMs provide the ideal tool to bring these sources together. We describe the methods used to assign values to selected words and phrases, and we introduce a method of bringing them to- gether to create a model for the classification of texts. In addition, several classes …

Content truncated to 500 words