A hybrid method for multi-class sentiment analysis of micro-blogs

With the development of social media, huge volumes of micro-blogs convey not only the factual information, but also the emotional status of individuals, which are crucial for understanding user behaviors in those micro-blogging systems. However, a micro-blog is typically very short and may contain rich sentiments other than the positive and negative, like the anxious, which brings great challenges to the so-called multi-class sentiment analysis. Although the model-based and lexicon-based methods are the two primary approaches extensively investigated and regularly used in this field, it is argued by some researchers that the model-based method provides poor results in multi-class analysis while the lexicon-based method is difficult to reflect the characteristics of short texts. In this paper, we propose a hybrid method for multi-class sentiment analysis of micro-blogs, which combines the model-based approach with the lexicon-based approach. Considering the effect of emoticons, we use emoticons and Naïve-Bayes classification to divide micro-blogs into three sentiments-positive, negative and neutral. After that, we use sentiment dictionaries to identify four negative sentiments-angry, sad, disgusted and anxious. We evaluate our algorithm on a real-life micro-blogging dataset collected from the popular Chinese micro-blogging site, Sina, and the results show that it is effective and efficient for timely sentiment analysis. Our method has been further applied to a Weibo User Profiling System and enabled the sentiment analysis of real-Time micro-blogs.