Topic generation for Chinese stocks: a cognitively motivated topic modeling method using social media data

With the explosive growth of user-generated data in social media websites such as Twitter and Weibo, a lot of research has been conducted on exploring the prediction power of social media data in financial market and discussing the correlation between the public mood in social media and the stock market price movement. Our previous research has demonstrated that the topic-based public
mood from Weibo can be used to predict the stock price movement in China. However, one of the most challenging problems in topic-based sentiment analysis is how to get the relevant topics about a stock. The relevant topics are also considered as concepts about a stock which can be used to build
the ontology of stock market for semantic computing and behavioral finance research. In this paper, motivated by the basic level concept in cognitive psychology, we present a novel method using Latent Dirichlet Allocation (LDA) to generate topics about a stock based on the social media data. The experimental results show that the proposed method is effective and better than other topic modeling methods. The topics generated by our method are more interpretable and could be used for topic-based sentiment analysis.