Human languages use coreference as a fundamental means of referring to a given person, place, concept or other entity across sentences and documents. For example, `John Smith' might be referred to in another sentence as `he' or `the CEO of GM,' or in another document with `John Smith Jr.' We have developed software which annotates text for this 'coreference' relation and will discuss a visualization technique based on node and link representations over a coreference annotated corpus. A major advantage of this approach is that users can meaningfully sift through large data collections using visual features of the representations in addition to more traditional text driven interfaces. I will present techniques for rapidly finding rare relationships or very common ones, new information detection about entities and focused speculative searching techniques.

Back to Workshop Schedule

Back to IMA "HOT TOPICS" Workshop: Text Mining