How Hierarchical Topics Evolve in Large Text Corpora.

摘要: Using a sequence of topic trees to organize documents is popular way represent hierarchical and evolving topics in text corpora. However, following the context remains difficult for users. To address this issue, we present an interactive visual analysis approach allow users progressively explore analyze complex evolutionary patterns topics. The key idea behind our exploit tree cut approximate each interactively modify cuts based on their interests. In particular, propose incremental algorithm with goal balancing 1) fitness smoothness between adjacent cuts; 2) historical new information related user A time-based visualization designed illustrate over time. preserve mental map, develop stable layout algorithm. As result, can quickly guide gain profound insights into We evaluate effectiveness proposed method Amazon's Mechanical Turk real-world news data. results show that are able successfully