This study aims to address the challenge of efficiently identifying appropriate citations in scientific literature, which has become increasingly difficult due to the rapid growth in the volume of scientific research. This study aims to enhance citation recommendation systems by incorporating both textual and bibliographic network information.
A novel attention-based encoder-decoder model is proposed, which integrates textual information within citation contexts and bibliographic network information outside of these contexts. The model uses a graph representation learning method to capture complex citation and co-author relationships within a bibliographic network. Additionally, academic pre-trained language models are utilized within an encoder-decoder framework to generate rich representations of citation contexts.
Extensive experiments on a real-world dataset demonstrate the effectiveness of the proposed method. Comparative experiments reveal that incorporating bibliographic network information significantly improves the performance of local citation recommendations.
This study is original in its approach to combining bibliographic network information with textual context for citation recommendation. By leveraging both types of information, the proposed method addresses a gap in existing citation recommendation systems, which often overlook the importance of external bibliographic relationships.
