Research On Academic Paper Recommendation Algorithm Based On Distributed Graph Model

With the advent of the era of big data,data processing solutions have emerged in various industries.The ability to process efficiently and store data is a basic requirement in the era of big data.The recommendation system has become an important technology.Academic paper is the main tool for scientific researchers to exchange academic ideas and research achievement,and which is an important research resource.In the field of academic paper recommendation,the recommendation system can dig out the potential user requirements of users through the analysis of the user's history records,feedback and other information.It also recommends related content to users in the area of requirements,and helps researchers find possible papers in tens of millions of papers.So that the user's process of searching for papers becomes more concise and convenient.The graph model abstractly expresses the real data in the form of a "graph"structure,and the graph algorithm is a calculation pattern driving on this structure.The graph model is an effective way to intuitively embody strong dependencies relationship between the paper data.This dissertation adopts the structural representation method of the graph model,proposes an academic paper recommendation algorithm based on the distributed graph model.First of all,according to the good structure information of the paper data,the algorithm extracts the feature respectively including title,abstract and citation relationship,and builds a paper relation graph model for each feature.The model can reasonably express the relationship between different structural types.Then,the model introduces user information to construct the user-paper two-layer graph model.Secondly,this dissertation proposes an evaluation method of paper quality.This method is an improvement of PageRank algorithm.It calculates the importance degree of each paper through the association degree of vertices in the two-layer graph,and it combined with SVD++ algorithm forms a hierarchical graph-based recommendation algorithm,named PRSVD++.This algorithm effectively solves the problem of low accuracy caused by sparsity.Finally,this dissertation uses the Spark distributed computing platform as the experimental environment,and it uses the graph computing framework named GraphX to implement the above algorithms.Compared with the standalone computer,the distributed platform of four nodes saves 28.1%of the running time.With the environmental advantages based on the memory computing,this platform can effectively solve the scalability problem of the algorithm for large-scale data.The experimental results show that the academic paper recommendation algorithm based on the distributed graph model are higher than the commonly used recommendation algorithm in the evaluation standard of recall,F measure and coverage.It verifies the rationality and effectiveness of hierarchical hybrid graph recommendation algorithm.