Exploring noise control strategies for UMLS-based query expansion in health and biomedical information retrieval

Abstract

In the past decades, ontology-based query expansion has been studied to improve health and biomedical information retrieval by many researchers, but the results of previous works are inconsistent. Query expansion with domain ontologies could introduce noise that degrades the retrieval performance, therefore noise control is the key to its success. In this paper, we explore three noise control strategies for UMLS-based query expansion. The first strategy is the adoption of a word-phrase hybrid retrieval model, and the other two strategies explored are expansion term weighting and term filtering. All the three strategies are implemented based on the Indri search engine and evaluated on two standard datasets, OHSUMED and TREC Genomic Track 2006. The experimental results indicate that the word-phrase hybrid retrieval model is superior to the word-based model and the pure phrase-based model, and beneficial to not only baseline retrieval but also query expansion. Expansion term weighting is an effective strategy to suppress term noise and improve retrieval performance. And expansion term filtering can also give some positive effects in most cases but is not as effective as the other two strategies. By combining the three strategies together, the best retrieval performances can be achieved on both datasets.