Mining the Relation between Sentiment Expression and Target Using Dependency of Words

Mining the Relation between Sentiment Expression and Target Using Dependency of Words Zhongchao Fei, Xuanjing Huang, and Lide Wu Department of Computer Science and Technology, Fudan University, Shanghai, 200433, China {zcfei, xjhuang, ldwu}@fudan.edu.cn Abstract. In some applications of opinion mining in text, it is important to distinguish what an author is talking about from the subjective stance towards the topic. Therefore, it needs to find the relation between sentiment expression and target. This paper proposes a novel method based on dependency grammars to mine the relation. In this method, the process of mining the relations is turned into a procedure of searching in the dependency tree of a sentence. The result of our experiments shows that word dependency relation based methods is more flexible and effective than some previous surface patterns based methods. Keywords: Opinion Mining, Dependency Grammars, Sentiment Analysis, Dependency of Words. 1 Introduction Opinion mining is also called sentiment analysis. It is to analyze whether an author expresses a favorable or unfavorable sentiment for a specific target. Now, it has been extensively used in text filtering, public opinion tracking, customer relation management, etc. [8],[9],[12],[19]. For example, during mining the opinion about a product of customers, we need to find what the customers remark on. It means we should not only extract the sentiment expressions, but also find the targets. Therefore, it needs to mine the relation between sentiment expression and target [9]. This paper will focus on it. The rest of the paper is arranged as following. Section 2 introduces some previous work. Section 3 provides how to gain subjective words. Section 4 introduces the dependency relation between words. Section 5 and section 6 put forward our method on mining the relation. Experiment and result are discussed in the last section. 2 Previous Work There is some previous work about opinion mining. Some previous work usually depend on the position between words [8],[9],[12],[17]. In some of the publications, n-gram was used [1],[15]. Turney (2002) utilized some phrase patterns such as RB/RBR/RBS+JJ+NN/NNS; JJ+NN/NNS+ Anything; RB+VB+ Anything. These patterns are composed of subjective adjectives, nouns, adverbs, etc. Nasukawa (2003) uses some patterns as V+Obj: admire somebody, JJ+Obj: crude oil. 257 V is the subjective verbs, JJ represents subjective adjectives. Obj represents the target. Nasukawa further defined some transfer verbs in the patterns, such as get, feel, etc, which may help to expression sentiment. Yi et al. (2003) utilized some patterns based on BNP. Popescu and Etzioni (2005) defined 10 kinds of rules to extract opinion patterns. For example, If ∃ (M, NP = f) po = M : (expensive) scanner, If ∃ (S = f, P, O) po = O : lamp has (problems), po=potential opinion, M=modifier, NP=noun phrase, S=subject, P=predicate, O=object. These rules point out which component of the patterns is the feature (target) and which is the subjective word. They also mentioned that these patterns are based on some dependency relations, but they did not depict them in detail. As we can see, the above methods acquire relations …

Content truncated to 500 words