Context-Aware Graph Neural Citation Recommendation Integrating Hierarchical Scholarly Semantics

Abstract

In order to improve the quality and accuracy of academic writing, automatic citation recommendation has recently been adopted. By combining deep contextual text models and graph-based scientific literature representations, this paper aims to build an effective system to enhance citation recommendations. In order to simultaneously understand the content and citation structure, a pair of semantic encoders and an adaptive attention mechanism are proposed. Experiments can use a wide range of academic datasets. The model performed excellently, ranking in the top 5 compared to unimodal and shallow fusion methods, with an accuracy of 55.2% and a mean average precision of 0.426. According to the results of the ablation study, language understanding and graph learning architectures should be used simultaneously to address the issue of ambiguous references and improve the reliability of retrieval. In other fields such as computer science and biomedicine, this method can also be applied. According to the findings, the new system will be more reliable and effective in providing citation recommendations. These findings may provide new directions for research and offer important support for other studies.