Evaluation Results -2
n
Baseline: pervious TRECVID 2005 system that uses heuristic
weights to link query and HLFs
n
Run1: Baseline + query matching without WordNet glosses
n
Run2: Run1 + WordNet glosses
n
Run3: Run2 + time-dependent similarity measure
0.106 (+1.9%)
Run1
0.110 (+5.8%)
Run2
0.113 (+8.6%)
Run3
0.104
Baseline
MAP
Models