Computing Sentiment Scores of Verb Phrases for Vietnamese

Computing Sentiment Scores of Verb Phrases for Vietnamese Thien Khai Tran1, Tuoi Thi Phan2 Faculty of Computer Science and Engineering, Ho Chi Minh City University of Technology Ho Chi Minh City, Vietnam 1 thientk@cse.hcmut.edu.vn, 2 tuoi@cse.hcmut.edu.vn Abstract. Sentiment analysis is an emerging research field. One of the major tasks of sentiment analysis is building sentiment lexicons and calculating their scores, which is an essential job that provides “material” for all sentiment analysis problems. In this paper, we propose a fuzzy language computation by taking linguistic context into account to provide an effective method for computing the sentiment polarity of verb phrases. The positive results, which come from an experimental period, will provide us with a basis from which to build an effective sentiment analysis system by making use of the contextual valence shifter. Keywords: sentiment lexicons; language computation; linguistic variable; fuzzy logic; fuzzy function; approximate reasoning 1. Introduction Sentiment analysis (or opinion mining) is a new research field, but it is an important area that attracts the attention of not only researchers but also businesses and organizations. Building sentiment lexicons is an essential task that provides “material” for all sentiment analysis levels: document-based, sentence-based, concept-based, and aspect-based. One of the biggest English sentiment lexicons is SentiWordNet [15]. It contains opinion terms extracted from WordNet [3] with a semi-supervised learning method and is available for research purposes. SenticNet [2] is a lexical resource used in concept-level sentiment analysis. It provides sentiment scores for 14,000 common sense concepts. To tackle the problem of mining verb expressions to identify opinions from customer reviews, there also have been a large number of works discovered the semantics of verbs and verb phrases. For example, Sokolova and Lapalme [13] incorporate semantic verb categories including verb past and continuous forms into features sets. Neviarouskaya et al. [9] built a rule-based approach to incorporate verb classes from VerbNet [12] to detect the sentiment orientation of sentences. For Vietnamese, Vu et al. [18] built VietSentiWordNet, which contains 1,000 words; it also includes syntactic rules for extracting sentiments from review sentences. Hong et al. [4] built an opinion dictionary for product domains based on a combination of a statistical method, a machine translation technique, and WordNet. Their work outperformed VietSentiWordNet. Recently, in 2016, Son et al. [14] built a Vietnamese opinion dictionary that contains five sub-dictionaries: verb, adjective, adverb, noun, and proposed features. The sub-dictionaries are based on the English emotional analysis approach and adapted to traditional Vietnamese language. The support vector The 2016 Conference on Computational Linguistics and Speech Processing ROCLING 2016, pp. 204-213  The Association for Computational Linguistics and Chinese Language Processing 204 machine classification technique was then used to identify the emotional content of the user’s message. However, the authors calculated the sum of the emotional values of the linguistic variables based on feelings. In this paper, based on Vietnamese linguistic characteristics and the fuzzy computation proposed by Zadeh [6,8,19], we present an effective method for computing the sentiment polarity of verb phrases. From this, we …

Content truncated to 500 words