TREC 2003 Genomics Track Experiments at UTA: Query Expansion with Predefinded High Frequency Terms.
Ari Pirkola, Erkka Leppänen · Text REtrieval Conference · 2003
We studied the effects of query expansion and query structure on retrieval performance. Two sets of words frequent in relevant documents for Genomics Track's training topics were collected, the first manually and the second automatically. The high frequency words collected and the names of organisms designated in the test topics, were used as expansion keys in gene name queries formed from the final test topics. The results indicated that Boolean structured queries expanded with automatically collected high frequency words and names of organisms performed considerably better than queries containing gene names only as keys. In the Boolean queries the expansion keys were categorized based on the aspects they represent in the documents discussing gene function. All the structured queries performed better than unstructured queries where each key contributed equally to document weights. In the structured queries gene names were assigned more weight than the expansion keys.