The School of Computer Science at the University of Windsor is pleased to present …
Comparative Study of Query Types on Query Refinement
PhD. Seminar by: Zahra Taherikhonakdar
Date: Wednesday, July 30th, 2025
Time: 10:00 AM
Location: Essex Hall, Room 122
Within a search session, users seek their information needs through iterative refinement of their queries, which is daunting. Neural-based query refinement methods aim to address this challenge via training on gold standard pairs of (original query à refined query), oblivious to the type of queries, be it factual like ‘table leafs or leaves’, navigational, e.g., ‘bbc good food recia’, or transactional such as ‘free play minecraft’, and, hence, fall short of finding refined versions for many original queries. In this paper, we bridge the gap by incorporating query types when generating refined queries. We fine-tune a conditional transformer, e.g., t5, to map an original query onto its relevant documents while conditioning on its type during training so that, during inference, the query type controls generating new reformulated queries within the same search intent for an unseen original query. Our experiments on the large-scale orcas-i-2m dataset across five query types demonstrated the synergistic effects of considering query types in generating more refined queries with better information retrieval efficacy. Specifically, considering query type has shown the best retrieval performance among navigational queries while yielding the highest number of refined queries in factual queries.
Internal Reader: Dr. Jianguo Lu
Internal Reader: Dr. Luis Rueda
External Reader: Dr. Tanja Collet-Najem
Advisor (s): Dr. Hossein Fani, Dr. Ziad Kobti
