Research on Online Topic Evolutionary Pattern Mining in Text Streams

Text Streams are a class of ubiquitous data that came in over time and are extraordinary large in scale that we often lose track of. Basically, text streams forms the fundamental source of information that can be used to detect semantic topic which individuals and organizations are interested in as well as detect burst events within communities. Thus, intelligent system that can automatically extract interesting temporal pattern from text streams is terribly needed; however, Evolutionary Pattern Mining is not well addressed in previous work. In this paper, we start a tentative research on topic evolutionary pattern mining system by discussing fully properties of a topic after formally definition, as well as proposing a common and formal framework in analyzing text streams. We also defined three basic tasks including (1) online topic Detection, (2) event evolution extraction and (3) topic property life cycle, and proposed three common mining algorithms respectively. Finally we exemplify the application of Evolutionary Pattern Mining and shows that interesting patterns can be extracted in newswire dataset.