FinancialFlow: Visual analytics of financial news based on hierarchical dirichlet process

Abstract

Topic evolution analysis is an important task in text mining to help people understand complex relationships between latent topics in large text collections and discover interesting topic evolution patterns. As most of the studies of topic evolution are made on open data set and few achievements are gained on textual visualization in the domain of finance, we propose a visual topic model named FinancialFlow to explore topic evolution patterns from financial news. Our model can be described as an integration of topic modeling algorithms and text visualization techniques, the key of which is to employ the hierarchical Dirichlet process to discover coherent topics in text data with temporal information and then visualize the process of how topics emerge, decline and develop in the form of flows. Finally, model evaluation and experimental results verify the feasibility and effectiveness of our proposed method in the real-world financial news data.