A Hybrid Neural Network Based on Multi-feature Fusion for Paper Recommendation

The purpose of academic paper recommendation is to provide researchers with papers of interest. Existing recommendation methods often takes the title or abstract as the paper's representation and learn based on text matching, ignoring the feature and structural information of the paper, resulting in poor semantic relevance to the recommendation results. A hybrid neural network paper recommendation method based on multi-feature fusion is proposed, which combines attention mechanism to learn deep semantic and structural features of papers to achieve classification and recommendation of papers in different academic fields. Experiments show that the method can better complete the classification and recommendation of papers, and has great accuracy compared with the traditional paper recommendation.