The School of Computer Science is pleased to present…
Log-Free Query-Type Modeling for Conditional Query Refinement in Information Retrieval
PhD Dissertation Defense by: Zahra Taherikhonakdar
Date: May 4
Time: 12:00 PM – 2:00 PM
Teams Link:
Abstract:
Query refinement plays a central role in information retrieval (IR) systems by bridging the gap between a user’s initial query and the underlying information need. However, real-world queries are often short, ambiguous and refinement strategies that apply uniform transformations across all queries frequently yield inconsistent or suboptimal outcomes.
This experiment systematically investigates the role of query types (user-intent categories) in query refinement and demonstrates that explicitly conditioning refinement processes on query type information leads to more principled, interpretable, and empirically effective refinement behavior. Retrieval effectiveness is assessed using standard IR metrics, including map, ndcg, and mrr, with statistical significance testing. To enable scalable and reproducible query type analysis, this dissertation further proposes an agent-based large language model (LLM) framework for automated query type annotation. The framework incorporates structured prompting, validation stages, and auditing mechanisms to improve annotation consistency and reliability, while operating in a log-free setting that avoids reliance on user interaction data.
Across datasets and retrieval models, the results demonstrate that intent-aware refinement yields consistent and statistically significant improvements over type-agnostic baselines, while maintaining robustness under domain variation. Overall, this dissertation presents an end-to-end framework that integrates query type modeling, agentic LLM-based annotation, and conditional refinement, providing empirical evidence and practical guidance for deploying scalable, type-aware query refinement in modern IR systems.
Keywords:
[Query Refinement, Query Type, Search Intent, Ai-agnet]
Doctoral Committee:
Internal Reader: Dr. Jianguo Lu
Internal Reader: Dr. Luis Rueda
External Reader: Dr. Tanja Collet-Najem
External Examiner: Dr. Samira Sadaoui
Advisor(s): Dr. Ziad Kobti
