Graph Convolutional Approaches to Scholarly Knowledge Discovery Across Disciplines

The rapid growth of scholarly publications across various domains presents both opportunities and challenges for academics seeking to synthesize knowledge. Conventional bibliometric studies can illuminate keyword trends, co-authorship, and citations, but they often fail to capture the complex relationships between writers, publications, and study subjects. Finding new trends and transdisciplinary knowledge is harder due to this restriction. Graph GCNs explain varied academic graphs in Convolutional Academic Knowledge Discovery (GCSKD). It addresses this issue. Nodes represent authors, publications, and keywords, while edges denote citations, co-authorships, and topic similarities in GCSKD. GCSKD utilizes both local and global neighborhood data to generate node embeddings that capture both structural and semantic information. These embeddings are used for link prediction, node classification, and community discovery to identify new research subjects, partnerships, and cross-disciplinary connections. It evaluates GCSKD on enormous bibliographic datasets from computer science, biomedical research, and social science. Experimental results show that GCSKD outperforms co-citation and embedding-based algorithms in research theme clustering and in predicting impactful publications and author collaborations. GCSKD provides a scalable and efficient framework for academics to discover new knowledge across fields. Collecting complex relational patterns in large academic networks enables the automated identification of trends across multiple scientific fields, helping researchers and institutions make informed decisions.