Detecting Manipulated Citations through Disturbed Node2Vec Embedding

This study presents an innovative approach for detecting potential citation manipulation in academic papers. Traditional methods often struggle to identify manipulated citations, undermining the credibility and reliability of scholarly publications. The proposed method introduces perturbations to a deep embedding citation graph model to reconstruct meaningful connections. A series of perturbations are applied within this network, involving randomly removing a fixed portion of edges. The objective is to recover trustworthy citations by considering alternative paths and assessing their reliability. Edges that exhibit more consistent behavior under these perturbations are identified as genuine, while those showing greater sensitivity are considered suspicious of manipulation. Numerical experiments provided on the Cora dataset validate the method’s effectiveness in identifying reliable citations as the core of the analyzed dataset and suspected anomaly citations.