Topic Extraction of Events on Social Media Using Reinforced Knowledge

摘要

The conventional topic models for topic extraction of events on social media are insufficient due to the data sparsity and the noise of microblog posts. The existing researches use word embeddings as prior knowledge to guide modeling or integrate conversation structures to enrich context. However, the shared context across a large number of events is ignored, which can be used as prior knowledge to reinforce coherent topic generation of each event. Thus, we propose a Reinforced Knowledge LDA for discovering topics of each event. It consists of three steps: (1) Running a topic model based on word embeddings and conversation structures to extract prior topics of each event; (2) Mining a set of reinforced knowledge sets from prior topics of all events automatically; (3) Using the reinforced knowledge sets to generate the final topics of every event. Experimental results on three real-word datasets which individually contain 50 events demonstrate the effectiveness of the proposed model and the reinforced knowledge.