Fusion of Entity Information from Textual Data Sources (e.g. HUMINT)

Fusing entity information based on discourse coreference will be explored. This research leads to capabilities such as change detection, the process of automatic notification when significant changes in a person or organization profile are detected. Change detection is an important element of the global awareness requirement and, in turn, supports Homeland Defense activities. One bottleneck in entity fusion is the insufficient support from coreference of entity mentions. About half of the locally extracted entity information cannot be successfully fused into profiles. The dilemma we face is this: if we indistinguishably use the results of coreference in fusion, the fused object is too noisy to be trustworthy. If we only use the high performance part of coreference, however, the fused object does not contain enough information to be valuable. A focused research effort is needed to advance the development of this critical enabling technology for IE fusion. The key is to develop a highly configurable and domain portable coreference capability which will demonstrate its value in generating high accuracy entity profiles. Our approach involves a combination of machine learning and grammatical processing.