Extracting Semantic Orientations of Phrases from Dictionary

Proceedings of NAACL HLT 2007, pages 292–299, Rochester, NY, April 2007. c⃝2007 Association for Computational Linguistics Extracting Semantic Orientations of Phrases from Dictionary Hiroya Takamura Precision and Intelligence Laboratory Tokyo Institute of Technology takamura@pi.titech.ac.jp Takashi Inui Integrated Research Institute Tokyo Institute of Technology inui@iri.titech.ac.jp Manabu Okumura Precision and Intelligence Laboratory Tokyo Institute of Technology oku@pi.titech.ac.jp Abstract We propose a method for extracting se- mantic orientations of phrases (pairs of an adjective and a noun): positive, negative, or neutral. Given an adjective, the seman- tic orientation classification of phrases can be reduced to the classification of words. We construct a lexical network by con- necting similar/related words. In the net- work, each node has one of the three ori- entation values and the neighboring nodes tend to have the same value. We adopt the Potts model for the probability model of the lexical network. For each adjec- tive, we estimate the states of the nodes, which indicate the semantic orientations of the adjective-noun pairs. Unlike ex- isting methods for phrase classification, the proposed method can classify phrases consisting of unseen words. We also pro- pose to use unlabeled data for a seed set of probability computation. Empirical evalu- ation shows the effectiveness of the pro- posed method. 1 Introduction Technology for affect analysis of texts has recently gained attention in both academic and industrial ar- eas. It can be applied to, for example, a survey of new products or a questionnaire analysis. Automatic sentiment analysis enables a fast and comprehensive investigation. The most fundamental step for sentiment analy- sis is to acquire the semantic orientations of words: positive or negative (desirable or undesirable). For example, the word “beautiful” is positive, while the word “dirty” is negative. Many researchers have de- veloped several methods for this purpose and ob- tained good results. One of the next problems to be solved is to acquire semantic orientations of phrases, or multi-term expressions, such as “high+risk” and “light+laptop-computer”. Indeed the semantic ori- entations of phrases depend on context just as the se- mantic orientations of words do, but we would like to obtain the orientations of phrases as basic units for sentiment analysis. We believe that we can use the obtained basic orientations of phrases for affect analysis of higher linguistic units such as sentences and documents. A computational model for the semantic orienta- tions of phrases has been proposed by Takamura et al. (2006). However, their method cannot deal with the words that did not appear in the training data. The purpose of this paper is to propose a method for extracting semantic orientations of phrases, which is applicable also to expressions consisting of unseen words. In our method, we regard this task as the noun classification problem for each adjective; the nouns that become respectively positive (negative, or neutral) when combined with a given adjective are distinguished from the other nouns. We create a lexical network with words being nodes, by con- necting two words if one of the two appears in the …

Content truncated to 500 words