Improved paper recommendation model incorporating knowledge graph embedding with IRGAN model

The exponential growth of published papers in digital libraries has made finding relevant papers increasingly challenging for researchers. Numerous citation recommendation models have been developed to mitigate this issue and assist researchers in identifying pertinent works. Unfortunately, many of these models struggle to generate high-quality recommendations because they fail to capture diverse relationship patterns and effectively model the multifaceted relationships present in citation networks. Additionally, these models lack robustness in learning the representations of research papers. To address these limitations, we propose an innovative model that integrates graph embeddings to better understand hidden relationships in bibliographic networks and model multi-fold relations using the MRotatE framework. Our model first leverages graphical feature embeddings and content-based representations through MRotatE and SPECTER document embedding to construct initial representations of both query and candidate papers. These representations are then used in a generator and discriminator, which undergo adversarial training, ultimately resulting in improved recommendation outcomes. We evaluated our model against several baselines using three real-world datasets, and the results demonstrate that our approach outperforms existing models. The code for this project is publicly available online. 1