Opinion Extraction based on Syntactic Pieces

Opinion Extraction based on Syntactic Pieces* Suguru Aoki and Kazuhide Yamamoto Nagaoka University of Technology, 1603-1, Kamitomioka, Nagaoka, Niigata 940-2188 Japan {aoki, ykaz}@nlp.nagaokaut.ac.jp Abstract. This paper addresses a task of opinion extraction from given documents and its positive/negative classification. We propose a sentence classification method using a notion of syntactic piece. Syntactic piece is a minimum unit of structure, and is used as an alternative processing unit of n-gram and whole tree structure. We compute its semantic orientation, and classify opinion sentences into positive or negative. We have conducted an experiment on more than 5000 opinion sentences of multiple domains, and have proven that our approach attains high performance at 91% precision. Keywords: syntactic piece, opinion extraction, sentence classification 1. Introduction One can easily disseminate information through the Internet, that include their personal opinions, such as reputation and dissatisfaction with products, complaints about services, and so on. Weblogs and message boards in particular have attracted a great deal of attention as a new information source, since they enable us to obtain subjective opinions easily. In order to automatically extract useful information from these sources, various approaches have been proposed. (Inui and Okumura, 2006) Researchers have been exploring techniques for classifying documents according to sentiment orientation, or positive/negative (p/n) in particular. Turney (2002) extracts phrases containing adjectives or adverbs, and determines their semantic orientation. Further, p/n of a document is judged by computing the average of the semantic orientations. The NEAR operator in the Alta Vista search engine is used in the method. If you search a query like A NEAR B in Alta Vista, the search engine shows pages containing within near words of each other. However, this operator is not provided in Japanese version. Wang and Araki (2007) extend Turney's method into Japanese by collecting and using a set of words that contribute significantly to p/n orientation. Fujimura et al. (2004) use corpora divided into p/n words, and statistically classify a document by extracting opinions. These methods provide the semantic orientation only with bag-of-words. A word, however, contains little and partial information. Therefore, we assert that the scope of the process needs to be expanded. Moreover, these methods do not identify reasons for p/n judgment, that is, expressions that cause semantic orientation. For a commercial application, this function is essential in marketing research. Besides these, other document classification approaches have also been proposed, which prepare a semantic orientation dictionary in advance. Tateishi et al. (2004) construct this * Copyright 2007 by Natural Language Processing Laboratory, Department of Electrical Engineering, Nagaoka University of Technology. 76 dictionary in advance by extracting opinion triplets that consist of `object name, attribute expression, evaluative expression’, and classify documents by using the triplets. The method extracts only expressions that appear in a definite pattern, and it is thus difficult to obtain satisfactory coverage and accuracy. Kobayashi et al. (2005) first extract a few opinion pairs of attribute and value with an anaphora resolution technique, and construct a semantic orientation dictionary of the pairs for a …

Content truncated to 500 words