A coupled linguistics/statistical technique for query structure classification and its application to Query Expansion

The retrieval effectiveness of Query Expansion (QE) is very much dependent on the ability to accurately identify and expand core concepts which are truly representative of the intended search goal. Two characteristics of natural language queries which hinder the performance of query expansion for information retrieval are query length and structure. The varying lengths of a query translate to the number of core concepts that may exist and the possibility of there being multiple query intents embedded within a single query. On the other hand, the structure of queries reveals the linguistic properties which allows for the determination of whether they take the form of well-formed sentences or are simply bags-of-words which in the strictest sense are a series of words with no obvious relations amongst them. Whilst query lengths are easily assessed, we propose a two-level automated classification technique consisting of linguistics based and statistical processing for query structure classification. The proposed method has revealed high levels of classification accuracy on TREC ad hoc test queries.

I. INTRODUCTION & RELATED WORK

Users express their information need through natural language queries which are not always well structured and may be semantically ambiguous. Depending on their familiarity with the search process, users construct queries which are both short and straight to the point or long-winded [7]. These queries may take the form of grammatically correct sentences or merely a group of keywords associated to their search goals. This basic variance in query formats motivates the need to analyze the structure and varying lengths of queries in order to decipher the intended search goal.