A Dynamic Conditional Random Field Based Framework for Sentence-Level Sentiment Analysis of Chinese Microblog

With the increasing popularity of social media, the Sentiment Analysis (SA) of the Microblog has raised as a new research topic. In this paper, we present WDCRF: a Word2vec and Dynamic Conditional Random Field (DCRF) based framework for Sentiment Analysis of Chinese Microblog. Our contributions include: firstly, to address drawbacks of Microblog message such as the length and Lexicon limitations, Word2vec technology is leveraged to enrich Microblog message so that each word individual is extended by its Top-k similar words. Secondly, DCRF model is utilized to combine and conduct the Subjectivity Classification and Polarity Classification simultaneously, while in existing works they are designed as independent and the relationship between two types of classifications is ignored. Moreover, the DCRF model considers not only the classification-level relationship but also the relationship between neighboring sentences. Finally, the experiments on real dataset collected from Sina and Tencent Weibo demonstrate that our WDCRF (Word2vec + DCRF) achieves much better than the state-of-the-art.