Due to the rapid expansion and heterogeneity of the data, it is a challenging task to discover the trends/paerns and relationships in the data, especially from a corpus of texts from published documents, news, and social media. In this paper, we introduce DycomDetector, a novel approach for topic modeling using community detections in dynamic networks. Our algorithm extracts the important terms/phrases, formulates a network of collocated terms, and then automatically refine the network on various features (such as term/relationship frequency, sudden changes in their time series, or vertex betweenness centralities) to reveal the structure/communities in the given network. These communities correspond to different hidden topics in the input texts. DycomDetector provides an intuitive interface and supports a range of interactive features, such as lensing or ltering, allowing users to quickly narrow down events of interest. We also demonstrate the applications of DycomDetector on several real-world datasets to evaluate its capabilities.
|State||Published - Aug 14 2017|