Research On Collaborative Filtering Recommendation Algorithm Integrating Ontology Semantics And User Attributes

Under the background of big data,the problem of information overload caused by massive information is increasingly prominent.How to efficiently and quickly find resources to meet users' needs has become an urgent problem to be solved.As an important tool to solve the above problems,recommendation system has been widely studied and applied in academia and industry.At present,how to provide more personalized Suggestions for users is the main research goal of recommendation algorithm.As one of the most classical algorithms in the recommendation field,collaborative filtering algorithm makes personalized recommendations for target users through user groups with similar preferences.However,with the increasing amount of data,the traditional collaborative filtering algorithm has some problems,such as data sparsity and the decrease of recommendation accuracy caused by the phenomenon of long tail.In order to solve the above problem,a collaborative filtering recommendation algorithm is proposed which integrates ontology semantics and user attributes.Firstly,according to the attribute information of items in the ontology database,the similarity between items is calculated by ontology semantic similarity method,and the project similarity matrix is constructed.Secondly,user attribute information,namely user interest degree and difference degree,is extracted from the user-project scoring matrix to construct user similarity matrix,and then user preference matrix is formed by integrating ontology semantics and user attributes.Finally,the prediction score of the user preference matrix is weighted to complete the top-n recommendation.In the experiment,Movielens data set was adopted,and the results showed that compared with the current popular recommendation algorithm,the collaborative filtering recommendation algorithm that integrates ontology semantics and user attributes has the lowest MAE value,and the accuracy rate is 71%.So it is superior to other recommendation algorithms in terms of integrity and novelty.The thesis has 20 figures,16 tables,and 68 references.