Sentiment Sentence Extraction Using a Hierarchical Directed Acyclic Graph Structure and a Bootstrap Approach

Sentiment Sentence Extraction Using a Hierarchical Directed Acyclic Graph Structure and a Bootstrap Approach * Kazutaka Shimada a, Daigo Hashimoto a, and Tsutomu Endo a a Department of Artificial Intelligence, Kyushu Institute of Technology 680-4 Iizuka Fukuoka 820-8502, Japan {shimada, d_hashimoto, endo}@pluto.ai.kyutech.ac.jp Abstract. As the World Wide Web rapidly grows, a huge number of online documents are easily accessible on the Web. We obtain a huge number of review documents that include user’s opinions for products. To classify the opinions is one of the hottest topics in natural language processing. In general, we need a large amount of training data for the classification process. However, construction of training data by hand is costly. The goal of our study is to construct a sentiment tagging tool for particular domains. In this paper, we propose a method of sentiment sentence extraction for the 1st step of the system. For the task, we use a Hierarchical Directed Acyclic Graph (HDAG) structure. We obtained high accuracy with the graph based approach. Furthermore, we apply a bootstrap approach to the sentiment sentence extraction process. The experimental result shows the effectiveness of the method. Keywords: Sentiment analysis, sentence extraction, similarity, HDAG. 1. Introduction As the World Wide Web rapidly grows, a huge number of online documents are easily accessible on the Web. Finding information relevant to user needs has become increasingly important. The most important information on the Web is usually contained in the text. We obtain a huge number of review documents that include user’s opinions for products. Buying products, users usually survey the product reviews. More precise and effective methods for evaluating the products are useful for users. To classify the opinions is one of the hottest topics in natural language processing. Many researchers have recently studied extraction and classification of opinions (Hatzivassiloglou and McKeown 1997, Kobayashi et al. 2005, Pang et al. 2002, Wiebe and Riloff 2005). There are many research areas for sentiment analysis; extraction of sentiment expressions, identification of sentiment polarity of sentences, classification of review documents and so on. The goal of our study is to easily construct a corpus for sentiment information for particular domains that users want. For the purpose we need to extract sentiment sentences from documents as the 1st step of the corpus construction. Extraction of sentiment expressions or sentiment sentences is one of the most important tasks in the sentiment analysis because classification tasks usually need a large amount of training data to generate a high accuracy classifier. There are several reports for classification of sentences (Kudo and Matsumoto 2004, Osajima et al .2005). However, the purpose of these studies is to classify sentences into positive and negative opinions. Our purpose in this paper is to classify sentences into opinions and non- * Copyright 2008 by Kazutaka Shimada, Daigo Hashimoto, and Tsutomu Endo. This research was partially supported by the Ministry of Education, Science, Sports and Culture, Grant-in-Aid for Scientific Research (C), 18500115, 2008. 341 22nd Pacific Asia Conference on Language, Information and Computation, pages …

Content truncated to 500 words