Automatic construction of domain-specific sentiment lexicon based on the semantics graph

Abstract

Sentiment lexicon, which is the basis of research on opinion mining and sentiment analysis, plays an important role in the field of natural language processing. The generalized sentiment lexicon lacks domain adaptability that does not adequately meet the sentiment analysis needs of the target domain, so automatically building a domain-specific sentiment lexicon is particularly important for specific domain of sentiment analysis. In this paper, a semi-supervised method of automatically constructing domain-specific sentiment lexicon based on corpus is proposed. The semantics graph is constructed by extracting the sentiment words as nodes and calculating the similarity of sentiment words as edge weights. A method of point-wise mutual information considering global information, local information and constraint information is proposed to calculate the similarity of sentiment words, which can more comprehensively and accurately reflect relevance of words in corpus. Sentiment seeds are obtained from corpus by using the degree of graph theory so as to have more domain characteristics and greater coverage. Experimental results on multiple datasets show that the proposed method achieves better results in constructing domain-specific sentiment lexicon.