A Sentiment Analysis Algorithm of Danmaku Based on Building a Mixed Fine-grained Sentiment Lexicon

The Danmaku is a form of instant video text commentary that reflects the viewer's sentiment orientation. Currently, most of sentiment analysis algorithms based on the sentiment lexicon are using manual construction of the lexicon. However, this kind of method has strong subjectivity and low accuracy, and there is no mature sentiment analysis algorithm for the linguistic characteristics of the Danmaku. In order to solve these problems, we propose a sentiment analysis algorithm of Danmaku based on the construction of a mixed fine-grained sentiment lexicon. This paper realizes the sentiment analysis in three parts: the construction of a basic lexicon, the construction of a fine-grained sentiment lexicon, and the calculation of sentiment values. The basic lexicon includes a stopword list with Danmakus' characteristics and a link-word lexicon, while the fine-grained sentiment lexicon is constructed by using the way of combining word embedding and word co-occurrence. The final experimental results show that our method proposed in this paper is effective in analyzing the sentiment of the Danmaku texts.