Research On Network Literature Recommendation Algorithm Based On Machine Learning

With the rapid development of the Internet,researchers are more and more free to upload and download documents,and more and more electronic documents appear in the network.However,how to retrieve the electronic literature more accurately and quickly has become an urgent problem in the current scientific research work.In order to provide more efficient and accurate services,this paper proposes an electronic literature recommendation algorithm based on machine learning,which introduces the community discovery into the electronic literature recommendation system,While providing more accurate personalized recommendation,it effectively improves the problem of data sparsity and cold start in collaborative filtering recommendation algorithm.In the research field of scientific research community discovery,this paper proposes an overlapping community discovery algorithm(CODA-BS algorithm)which introduces community degree subordination bias.In each iteration,CODA-BS algorithm uses information entropy to calculate node threshold,uses bias function to optimize tag membership coefficient,and uses tag propagation rules to filter tags and normalize membership coefficient,so as to detect better communities.Through the validation on the benchmark data set in the field of community discovery and the academic resources and social platform data set of China Association for science and technology,from the experimental data analysis results,CODA-BS algorithm has higher accuracy and stability than the classic overlapping community discovery algorithm based on label propagation(COPRA algorithm).In the stage of E-literature recommendation,the results of community discovery are introduced into the subject,and certain strategies are adopted to make up for the user rating matrix.Firstly,the domain membership degree of community discovery is used to fill in the user domain score matrix,and then the user similarity is calculated according to the Pearson coefficient to improve the user based collaborative filtering recommendation;The collaborative filtering recommendation based on literature uses the conditions of literature membership domain and author's domain membership degree,and uses certain strategies to transform it into user literature score table,then calculates the literature similarity and makes recommendation,so as to complete the collaborative filtering recommendation integrating community discovery;Finally,a new idea is proposed: Several recommendation models are switched and mixed.According to the different data sources,the model will choose different recommendation algorithms for training,and the most appropriate recommendation results can be obtained through this switching method.Experimental results show that collaborative filtering recommendation and switching hybrid recommendation based on community discovery have better experimental results than content-based recommendation and collaborative filtering recommendation in the field of scientific research community.