Collab identifier: a prototype for uncovering research collaboration themes using BERTopic

Abstract

This study introduces Collab Identifier, a prototype designed to identify potential research collaboration opportunities between institutions using scholarly publication data. The study explores how integrating Natural Language Processing (NLP) and Library and Information Science (LIS) approaches may support the discovery of thematic collaborations in increasingly interdisciplinary research environments. Publication abstracts indexed in Scopus from two institutions were analyzed using BERTopic, a contextual topic modeling approach. Institutional topic models were subsequently compared using cosine similarity to identify thematic alignment and specialization. The resulting topic labels and collaboration recommendations were validated through expert review, including a semantic assessment of AI-assisted labels and proposed collaboration themes. BERTopic achieved higher topic coherence scores than the Latent Dirichlet Allocation (LDA) baseline across both institutional corpora. Similarity-based recommendations remained stable around the baseline threshold but contracted under stricter thresholds, while expert reviewers generally perceived AI-assisted topic labels and collaboration recommendations as meaningful and practically relevant. The findings further indicate that semantic similarity alone is insufficient to ensure conceptually meaningful collaboration, highlighting the importance of expert validation. This study extends conventional bibliometric approaches by integrating contextual topic modeling and expert validation to identify prospective opportunities for institutional collaboration. Unlike profile-centric systems such as VIVO, Collab Identifier focuses on uncovering latent thematic alignment directly from publication abstracts and is intended to complement, rather than replace, existing research discovery platforms.