Cyber Times International Journal of Technology & Management

Millions of users to share their opinions on Twitter, making it a valuable platform for the monitoring and analysis of public opinion. This monitoring and analysis can provide essential information for decision making in various fields. Therefore, it has attracted attention both in academia and industry. Previous research mainly focused on modeling and monitoring of public opinion. In this work, It goes a step further to interpret the sense of variations. It is observed that emerging issues (leading subjects named) in climate change periods are closely linked to the real reasons behind the changes. Based on this observation, It propose a Dirichlet Allocation (LDA) latent base model, original and basic LDA (FB-LDA) to distill the subjects leading and filter longstanding substantive issues. These key issues can give interpretations of sentiment variations. To further improve the readability of mined reasons, It select the most representative tweets leading topics and develop another model called generative candidates Reason and Background LDA (RCB-LDA) to classify them according to their "popularity" in the period of variation. Experimental results show that our methods can actually find prominent subjects and candidates because of rank. The proposed model can also be applied to other tasks such as research about differences between two sets of documents.