Abstract Summary
The search for knowledgeable people is essential in academic settings, supporting the identification of suitable supervisors, reviewers, and collaborators. However, existing systems primarily focus on topical relevance while overlooking contextual factors such as supervision capacity, collaboration intent, and recency of activity. This research aims to design context-aware ranking models that incorporate behavioral and relational factors to improve expert identification across academic tasks. To measure progress, we conducted a systematic comparison of expert-finding algorithms on benchmark datasets, revealing that no single method consistently outperforms others and that effectiveness varies across query types and text elements. To address the limitations of existing benchmarks, we introduce the Find an Expert (FaE) dataset from the University of Melbourne, which integrates rich academic profiles, authentic interaction logs, and system-generated search results. Preliminary analysis shows strong position bias and multi-candidate exploration patterns that distinguish expert search from general web search. Building on these insights, the next stage of this research focuses on developing context-aware models that leverage behavioral features to enhance ranking effectiveness across diverse expert-seeking scenarios.
