Query Expansion in Resource-Scarce Languages: A Multilingual Framework Utilizing Document Structure

Retrievals in response to queries to search engines in resource-scarce languages often produce no results, which annoys the user. In such cases, at least partially relevant documents must be retrieved. We propose a novel multilingual framework, MultiStructPRF, which expands the query with related terms by (i) using a resource-rich assisting language and (ii) giving varied importance to the expansion terms depending on their position of occurrence in the document. Our system uses the help of an assisting language to expand the query in order to improve system recall. We propose a systematic expansion model for weighting the expansion terms coming from different parts of the document. To combine the expansion terms from query language and assisting language, we propose a heuristics-based fusion model. Our experimental results show an improvement over other PRF techniques in both precision and recall for multiple resource-scarce languages like Marathi, Bengali, Odia, Finnish, and the like. We study the effect of different assisting languages on precision and recall for multiple query languages. Our experiments reveal an interesting fact: Precision is positively correlated with the typological closeness of query language and assisting language, whereas recall is positively correlated with the resource richness of the assisting language.