An analytical deep learning framework for interpretable author similarity and research collaboration

This study proposes a hybrid framework combining Deep Learning (DL) and Fuzzy Multi-Criteria Decision-Making (MCDM) to evaluate author similarity and enhance collaboration in bioinformatics research. The model integrates structured expert judgment through the Analytic Hierarchy Process (AHP) and Fuzzy Delphi Method (FDM) with domain-specific Natural Language Processing (NLP) using Bidirectional Encoder Representations from Transformers for Biomedical Text Mining (BioBERT) embeddings. Key evaluation criteria include citation similarity (highest weight), content similarity (titles, abstracts, keywords), co-authorship history, institutional affiliation, field alignment, and journal venue overlap. These factors are weighted by consensus and aggregated into a transparent scoring formula. Validated on biomedical informatics data, the framework demonstrates strong performance in identifying Potentially Associated Author (PAA), achieving 0.89 precision@10 and 0.82 ranking stability (Jaccard Index). The approach uniquely bridges semantic analysis (via BioBERT) and expert-driven evaluation, offering interpretable, adaptable author recommendations tailored to bioinformatics’ interdisciplinary challenges. • Propose a hybrid framework combining deep learning and fuzzy decision-making for author similarity. • Demonstrate superior accuracy in identifying research collaborators compared to traditional approaches. • Provide a transparent and interpretable scoring method for research partner recommendations. • Validate the framework on large-scale biomedical data to ensure real-world applicability. • Offer a modular design adaptable to multiple interdisciplinary research fields.