FindCoResearcher: Retrieval-based Collaborative Researcher Recommendation System

Researchers primarily depend on personal networks or manual exploration when searching for collaborative researchers. This approach presents significant limitations in identifying researchers suitable for their research topics. While Information Retrieval (IR) approaches can address these limitations, two major challenges arise: datasets with limited expression diversity and the gap between realistic scenarios and IR research. We propose FindCoResearcher, a collaborative researcher recommendation system that incorporates a query generation methodology for increasing expression diversity and a new evaluation metric for bridging realistic scenarios. To ensure query diversity, we construct query sets for each passage by diversifying query styles and specificity levels based on augmented passages. We introduce a researcher-unit Top-k Accuracy evaluation approach that better reflects realistic scenarios. We fine-tuned a dense encoder BGE-M3 using DPR achieving superior researcher search performance. FindCoResearcher is expected to promote industry-academia collaboration and expand their collaborative networks.