DETECTING CITATION ANOMALIES ACROSS SCHOLARLY DATABASES USING GRAPH NEURAL NETWORKS

Purpose - This study proposes a Graph Neural Network (GNN)-based framework for detecting citation anomalies across Scopus, Web of Science, and Google Scholar to improve the reliability, transparency, and accuracy of bibliometric evaluation. Methodology - A qualitative research approach was adopted through an extensive review of citation analysis, graph learning, bibliometric research, and scholarly database evaluation. A conceptual Graph Neural Network framework is proposed to model citation relationships and identify anomalous citation patterns. Findings - The proposed framework demonstrates that Graph Neural Networks effectively capture complex citation relationships and identify anomalous citation behavior, duplicate records, inconsistent citation links, and unusual citation clusters across multiple scholarly databases. Practical Implications - The framework assists researchers, publishers, universities, librarians, and research evaluation agencies in improving citation quality, validating bibliometric data, and supporting reliable scholarly assessment. Originality - This study introduces a Graph Neural Network-based framework for citation anomaly detection across multiple scholarly databases, providing a novel graph-learning approach for enhancing bibliometric reliability and research evaluation.