Extracting Time Critical Information from Dynamic Text

We propose a novel approach to extracting domain-specific, time-critical information from the text of online chat. Tradition Information Extraction (IE) approaches are unsuitable for analyzing chat streams for many reasons: the text is “dirty” (containing typos, misspellings, sporadic use of case, etc.), messages are often fragmented and refer implicitly to previous messages and shared knowledge, messages from multiple topics are interleaved, etc. Our Chat-IE system will use a collection of techniques to process a chat stream that is being used in support of an ongoing activity. Exploiting the context of the ongoing activity will be crucial to extraction effectiveness, as will teasing out the individual dialogues that structure the chat stream. The IE process must also be modified in order to support extraction from those dialogues. In addition, the results of the entire process must be analyzed to determine its accuracy, criticality, completeness, and veracity. Our Phase I prototype will provide a solid foundation for the complete implementation of Chat-IE in Phase II and its commercialization.