US 7,577,654 B2
Systems and methods for new event detection
Thorsten H. Brants, Palo Alto, Calif. (US); Francine R. Chen, Menlo Park, Calif. (US); and Ayman O. Farahat, San Francisco, Calif. (US)
Assigned to Palo Alto Research Center Incorporated, Palo Alto, Calif. (US)
Filed on Jul. 25, 2003, as Appl. No. 10/626,856.
Prior Publication US 2005/0021324 A1, Jan. 27, 2005
Int. Cl. G06F 7/00 (2006.01); G06F 17/30 (2006.01)
U.S. Cl. 707—6 14 Claims
OG exemplary drawing
 
1. A computer-implemented method of detecting new events comprising the steps of:
determining at least one story characteristic based on an average story similarity story characteristic and a same event-same source story characteristic;
determining a source-identified story corpus, each story associated with at least one event;
determining a source-identified new story associated with at least one event;
determining story-pairs based on the source-identified new-story and each story in the source-identified story corpus;
determining at least one inter-story similarity metric for the story-pairs; wherein the inter-story similarity metrics are comprised of at least one story frequency model and at least one story characteristic frequency model combined using terms weights; and wherein an event frequency is determined based on term t and rule of interpretation (ROI) category rmax from the formula:

OG Complex Work Unit Drawing
wherein r is an ROI category, R is the set of all possible ROIs, ef(rt) is the frequency of term t in ROI category r:
determining at least one adjustment to the inter-story similarity metrics based on at least one story characteristic; and
outputting a new story event indicator if the event associated with the new story is of a determined similarity or dissimilarity to the events associated with the source-identified story corpus based on the inter-similarity metrics and adjustments.