Research On Paper Recommendation Based On HIN Representation Learning

Faced with the massive amount of academic information on the Internet,how can researchers quickly and accurately find academic articles that match their research interests and research themes is an urgent problem to be solved,and recommendation technology is a typical strategy to solve this problem.Academic network as a heterogeneous information network,in which there are many types of nodes and links.How to make use of the information of researchers and papers in the recommendation at the same time,preserving the integrity of information in the network and fully learning the relationship between nodes is a challenging problem for paper recommendation.However,most of the current mainstream recommendation technologies only consider text information or citation information,and do not fully consider the staggered relational information in the academic network.In order to solve this problem,this paper introduces heterogeneous information network representation learning to carry out research,comprehensively considers the information of papers,authors,terms and conference type nodes in the academic network and the semantic information of different relationship types.In this paper,two novel presentation learning strategies are proposed to solve the problem of information representation learning in academic networks.The main contributions of this paper are as follows:(1)Aiming at the characteristics of diverse types of nodes and different path influences between nodes in academic networks,this paper proposes a representation learning method for heterogeneous information networks that integrates node weights and path weights.According to node neighborhoods and different types of metapaths,calculate dual weight information,highlighting the importance of different nodes under the metapath and the importance of different types of metapaths,improves the paper recommendation performance.(2)Aiming at the influence of the information in the heterogeneous information network cannot be fully captured on the paper recommendation performance,this paper designs a heterogeneous information network learning method embedded in the sub-network,which constructs the node feature sub-network and the metapath sub-network.And through the designed node relationship enhances attention unit fusion of the information in the two sub-networks,avoid the loss of information,modeling the relationship in the academic network,and improve the quality of the paper recommendation service.(3)In this paper,two proposed representation learning strategies are tested on several common datasets.After sufficient experiments,the proposed heterogeneous network representation learning method is compared with the baseline model,which proves that the heterogeneous information network representation learning method proposed in this paper can better integrate into the paper recommendation,thereby improving the performance of the recommendation.