Centering Resonance Analysis: A superior data mining algorithm for textual data streams

An increasing amount of information comes in the form of streaming text: news media, email, and even human conversation. Creating insight from streaming text has been challenging though, because current knowledge discovery systems depend on expert analysts to move from simple data to actionable knowledge. In order for such systems to be used outside a small cadre of experts, they must incorporate deep analytics, which are information rich and diagnostic in nature, and when combined with domain knowledge, point to action. In Phase I, Crawdad Technologies demonstrated the superior text mining performance and scalability of its Centering Resonance Analysis (CRA). A topic tracking system based on CRA had up to five times better precision than ones based on existing technologies. In Phase II, Crawdad will develop deep analytics that capture (a) the temporal, dynamic nature of streaming text, and (b) the social dynamic that underlies many streaming text applications. We will create Dynamic CRA (DCRA) and Multi-Agent DCRA to model streaming text in applications such as communication monitoring, news streams, team and group conversations, and customer service interactions. The use of DCRA and Multi-Agent DCRA will be tested in an actual news streaming environment.