Modeling Topic Evolution in Social Media Short Texts

Social media short texts like tweets and instant messages provide a lot of valuable information about the hot topics and public opinion. Detecting and tracking topics from these online contents can help people grasp the essential information and its evolution and facilitate many applications. Topic evolution models built based on LDA need to set the topic number manually, which could not change during different time periods and could not be adjusted based on the contents. The nonparametric topic evolution models do not perform very well on short texts due to the data sparsity problem. So in this paper, we propose a nonparametric topic evolution model for short texts. The model uses the recurrent Chinese restaurant process as the prior distribution of topic proportions. Combining it with word co-occurrence modeling, we construct a topic evolution model which is suitable for social media short texts. We carry out experimental studies on twitter dataset. The results show that our method outperforms the baseline methods and could monitor the topic evolution in social media short texts effectively.