Design And Implementation Of Paper Recommendation System Based On Graph Model

With the publication of a large number of academic papers,how researchers found academic papers that met their needs in a wide range of academic papers had became a thorny issue that haunted them.The recommendation of academic papers had became an effective way to solve the above problems.Most of the existing methods of academic paper recommendation used reference information,content information to recommend,but for academic papers with multiple types of feature information,the use of this method alone will lead to poor recommended results.Based on this,this study on the integration of multi-type feature information of papers by constructing graph model,a paper recommendation algorithm based on graph model was designed to solve the problem of information overload of current academic papers.The main research work of this paper is as follows:(1)Construction of Graph Model Based on Multi-type Feature Information of Papers.Aiming at the problem that multi-type feature information cannot be expressed in academic papers,a four-layer graph model was studied and constructed,which integrated the paper,author,subject and keyword information and the relationship between them.Considering the problem of data sparsity caused by papers with less relation in paper citation in graph model,the study used Attention-based double GRU to encode the paper semantically,the paper was understood from the semantic aspect,and the similarity relationship between the semantics of the paper was incorporated into the reference matrix of the paper.The experimental results showed that the understanding of the paper from the semantic point of the paper can find more similar academic papers than the target paper.(2)Design of paper recommendation alogrithms based on graph model.In this paper,the graph model and the reboot random walk algorithm were combined to design a paper recommendation algorithm based on graph model to improve the accuracy of the recommended paper.Experimental results showed that,on the AAN dataset,compared with QS-PageRank algorithm,LDA algorithm,Link-PLSA-LDA algorithm,RTM algorithm and PAWRW algorithm,the recommendation algorithm proposed in this paper,an average increased of 6.37% on MAP,an average increased of 5.26% on the MRR,an average increased of 12.15% on the Recall.(3)Design and implementation of paper recommendation system.Aiming at the problem that the information utilization rate of the existing paper recommendation method is not high,this paper proposed and implemented the paper recommendation system by used the Java language,Spring Boot framework and CSS,JS and other technologies in combination with the proposed method.