A novel framework with ComMAND: A combined method for author name disambiguation

Abstract

Author Name Disambiguation (AND) in digital bibliographic repositories is a persistent challenge due to homonyms and synonyms, compromising information retrieval and database integrity. This work presents a novel framework with a Combined Method for Author Name Disambiguation (ComMAND) that integrates transfer learning with SciBERT, Graph Convolutional Network (GCN), and Graph-enhanced Hierarchical Agglomerative Clustering (GHAC) to enhance AND performance. The framework includes a Graphical User Interface (GUI), allowing users to load datasets, execute AND tasks, and visualize results without requiring programming knowledge. By semantically analyzing document content and leveraging graph-based relationships, our approach achieves higher precision in identifying unique authors. Experimental results on AMiner-12, AMiner-18, and DBLP validate the effectiveness of the framework. Considering the DBLP dataset, which contains extensive ambiguous name references (679), the results show the highest F1 of 0.869 and K-metric of 0.972 compared to the baseline works, with improvements ranging from 1.1% to 33.6% over baseline works. These findings highlight the effectiveness of combining machine learning, graph-based techniques, and clustering for large-scale AND tasks.