Revisiting Again Document Length Hypotheses TREC-2004 Genomics Track Experiments at Patolis
Sumio Fujita · 2004
The TREC-2004 Genomics track evaluation experiments at Patolis Corporation are described with a focus on the document length issues in different retrieval models such as TF*IDF or probabilistic language modeling approaches. In the genomics ad hoc retrieval task, combination of pseudo-relevance feedback and reference database feedback is applied. For the triage sub-task, we trained a SVM classifier using leave-one-out-cross-validation, and calibrated parameters to be optimal against the training set.