Interactive query suggestion in iPubMed

Interactive Query Suggestion presents semantically related query suggestions to a given user query and on the fly revises them for more relevant suggestions, as the user is typing the query keyword by keyword. This aids the user in query formulation especially when the underlying data has vocabulary problems like near-synonyms and term ambiguities. iPubMed is a medical search engine that allows keyword search for publications in the MEDLINE database. It supports an interactive search-as-you-type interface; and also allows approximate matching of keywords. The focus of the work is (1) Augmenting the iPubMed interface with a query suggestion feature, and (2) developing index structures and algorithms for realizing this feature. We use an existing domain-specific ontology to build sets of related queries and use a trie-based index structure to efficiently compute ranked suggestions. We have developed a prototype of this feature using the MeSH dataset and evaluated it on a real one-day query log from a popular medical search engine. Also, we present the results of a usability study, performed on a prototype integration of this feature with the iPubMed interface.