A Heterogeneous Graph Neural Network-based Topic Model for Research Trend Analysis

Recent advancements in big data and artificial intelligence (AI) technologies have highlighted the need for more sophisticated methodologies to analyze large-scale textual data and uncover meaningful research trends. While traditional topic modeling techniques have been widely used for trend analysis, they often face challenges in maintaining clarity and coherence, particularly in datasets with complex relationships and diverse metadata. This study proposes a new heterogeneous graph neural network-based topic modeling framework (HeteroGTM) designed to address these limitations. By modeling the intricate relationships between papers, authors, and keywords as a heterogeneous graph, HeteroGTM integrates graph neural networks with k-means clustering to extract consistent and interpretable topics. To validate its effectiveness, HeteroGTM was applied to federated learning-related academic data, providing a comprehensive analysis of recent research trends in this field. The model demonstrated superior performance compared to traditional methods, achieving higher coherence and modularity scores. Additionally, HeteroGTM's versatility was confirmed through its successful application to other domains, including explainable AI and edge AI, where it consistently outperformed existing techniques. The proposed approach enhances the performances of topic modeling and paves the way for future research in data-driven academic network analysis.