Scientific Literature Recommendation with Knowledge Graph and Collaborative Perception
Year: 2026
Type: book-chapter
Abstract: Accurately identifying the association between user interests and candidate literature has always been crucial for scientific literature recommendation. However, due to the diversity of user interests and literature topics, there are many noisy associations between user interests and candidate literature. The problem of eliminating these spurious associations has become a critical area of investigation. Collaborative perception, as a selective information-fusion technique, can effectively address diverse user interests and the multi-topic nature of the literature. In the context of scientific literature recommendation systems, if the document context on which the model focuses is unrelated to the collaboratively perceived document, it can generate ambiguous or noisy attention, negatively impacting the results. To address this problem, we propose an adaptive collaborative perception network using co-attention and gating mechanisms. A scientific literature recommendation model is constructed by integrating a knowledge graph. Experiments conducted on the CiteULike dataset demonstrate that the proposed model significantly outperforms baseline models in terms of AUC (0.874), MRR (0.582), and nDCG@10 (0.849). The model uses collaborative perception to effectively reduces noise and enhances the accuracy of recommendations. The model provides an innovative solution for scientific literature recommendation by effectively addressing the challenge of noise in user interests and diversity of document topics.
Authors: Li, Jie, Zhang, Guobiao
Institutions: Shandong University, Soochow University, Soochow University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7213605477