Topic Derivation in Weibo Through Both Interactions and Content

Social media represented by Weibo has become one of the most important platforms, covering all kinds of topics for information dissemination and day-to-day communications. As a result, topic derivation in Weibo can support various applications scenarios, including sentiment analysis, opinion controlling, market forecasting, etc. As traditional topic derivation in Weibo is mainly based on the short text of a weibo post, these methods usually encounter the data sparsity problem. To solve this problem, we find that both content and interactions can help improve the quality of topic derivation in Weibo. Thus, this paper proposed a method that additionally takes three typical interactions into features: mentioning, forwarding and the topic tags. The proposed method clusters the weibo posts and identifies the representative terms for each topic by matrix factorization technique. Our experimental results show that the proposed method performs better than advanced baseline methods in both topic clustering accuracy and keywords extraction.