New Event Detection at UMass Amherst Giridhar Kumaran and James Allan
CIIR, UMass Amherst2 Preprocessing Lemur Toolkit for tokenization, stopping, k-stemming BBN Identifinder™ for extracting named entities
CIIR, UMass Amherst3 Systems fielded Submitted four systems Didn’t include last year’s system Classification according to LDC categories and term – pruning Didn’t work on exclusively NW story corpus
CIIR, UMass Amherst4 Primary system – UMass1 Utility of named entities acknowledged Failure analysis indicates Large number of old stories have low confidence score (false alarms) Conflict with new story scores Reasons Stories on multiple topics Diffuse topics Varying document lengths
CIIR, UMass Amherst5 Primary system – UMass1 Focus Identify old stories better – affects cost Clue Most old stories get low confidence scores as topics linked by only named entities (large number) only non-named entities (few)
CIIR, UMass Amherst6 Primary system – UMass1 Approach Look at the set of closest matching stories If consistently high named entity or non-named entity match modify confidence score
CIIR, UMass Amherst7 Primary system – UMass1 Procedure Double original confidence score if less than a threshold Gradually reduce score towards original score if set of closest stories match neither named entities nor non-named entities
CIIR, UMass Amherst8 UMass1 – Examples from TDT3 Russian Financial Crisis - Old Story APW AllSimNESimnoNESim APW APW APW APW APW APW
CIIR, UMass Amherst9 UMass1 – Examples from TDT3 Russian Financial Crisis - Old Story APW AllSimNESimnoNESim APW APW APW APW APW APW
CIIR, UMass Amherst10 UMass1 – Examples from TDT3 Russian Financial Crisis - Old Story APW AllSimNESimnoNESim APW APW APW APW APW APW Threshold = 0.1
CIIR, UMass Amherst11 UMass1 – Examples from TDT3 Russian Financial Crisis - Old Story APW AllSimNESimnoNESim APW APW APW APW APW APW Threshold = 0.1
CIIR, UMass Amherst12 UMass1 – Examples from TDT3 Russian Financial Crisis - Old Story APW AllSimNESimnoNESim APW * APW APW APW APW APW Threshold = 0.1
CIIR, UMass Amherst13 UMass1 – Examples from TDT3 Thai Airbus Crash - New Story APW AllSimNESimnoNESim APW * APW APW APW APW APW
CIIR, UMass Amherst14 UMass1 on TDT3
CIIR, UMass Amherst15 UMass1 on TDT3
CIIR, UMass Amherst16 UMass2 Basic vector space model system Compare with all preceding stories Return highest cosine match
CIIR, UMass Amherst17 UMass3 Same model as UMass2 TDT5 – Very large collection Practical system Compare with a maximum of stories with highest coordination match Faster
CIIR, UMass Amherst18 UMass4 Similar to UMass1 Rationale is the same Consider top five matches Use different formula for modifying confidence score
CIIR, UMass Amherst19 Performance Summary System Topic weighted min. cost (TDT5) Topic weighted min. cost (TDT4) UMass1 – Modify confidence score based on evidence UMass2 – Basic vector space model UMass3 – UMass2 + restriction on number of documents compared with UMass4 – UMass1 with different formula
CIIR, UMass Amherst20 Summary Basic vector space model did the best Restricting number of stories to be compared with Improved system speed Didn’t improve performance Primary system did extremely well on training data, but failed on TDT5