Query Intention Acquisition: A Case Study on Automatically Inferring Structured Queries

The process of acquiring the user’s intentions is an important phase in the querying process. The identification of their intentions enables the selection of appropriate retrieval strategies. In this paper, we first outline this cross-section work in contextual Information Retrieval. We then focus on one particular type of intention: the syntactic and semantic types associated with a query term. We present a case study using the email search task of the TREC Enterprise Track. We build and analyze a data set of query intentions linked to the email’s structure, and then attempt to automatically infer structured queries and study the affect that ambiguity of queries and the difficulty of inferring them has on various retrieval models (structured and unstructured). Our study reveals that predicting the intentions is a hard problem due to the inherent uncertainty within the querying process. We also show that automatically inferred queries do not outperform other types of structured retrieval models, because they are not robust enough to handle the ambiguity nor reliable enough to be accurately inferred.