An Unsupervised Microblog Emotion Dictionary Construction Method and Its Application on Sentiment Analysis

摘要

Emotion lexicon plays an important role in many lexicon based sentiment analysis. A lot of work relating to emotion lexicon construction has been done, but most of them turn out to be time-consuming or laborious and cannot update in time. This paper proposes an automatic approach to construct Sina Weibo (a Chinese Twitter like microblog) emotion dictionary (Weibo Dic) by exploiting word2vec tool based on HowNet sentiment dictionary from a large-scale corpus. Firstly, we extract and classify emoticons contained in the corpus, and the resulted emoticons will serve as seeds in our approach. Secondly, we will make use of the word2vec tool to build a cluster model based on five million pieces of text after word segmentation. Thirdly, since we got the emoticons as seeds and the cluster model, we propose an iterative method to construct Weibo Dic. Finally, we analysis the Weibo Dic and verify the effect on sentiment analysis of Sina Weibo. The high precision shows that the effectiveness of the Weibo Dic is satisfactory.