Research And Implementation Of Recommendation Algorithm Optimization Based On User Similarity

With the rapid development of computers and the change of network technology,the Internet has gradually entered people's daily lives and completely changed the way people access information.A lot of network information can provide a lot of value,but most consumers will be plagued by various choices,and this huge information is prone to "information overload",can not be used more effectively,and will make It is getting harder and harder for users to find relevant information.At this time,the status of the recommendation system is highlighted.The recommendation system(RS)has proven to be an effective e-commerce tool that provides personalized recommendation technology to help users select attractive products.So far,the recommendation system has successfully found applications in e-commerce,such as books recommended at amazon.com,movies recommended at Netflix.com,and so on.Among many recommended technologies,collaborative filtering(CF)is the most widely used and most successful method in RS,but there are still some shortcomings.For example,the problem of data sparsity still plagues the accuracy of recommended performance.Therefore,based on the traditional collaborative filtering recommendation technology,this thesis conducts in-depth research and analysis on the problems that arise.The main research contents are as follows:(1)This thesis introduces the historical development of collaborative filtering algorithms and the development results in recent years.It also introduces relevant recommendation techniques,including several common recommendation algorithms.Several problems existing in the recommendation system are analyzed.Based on the problem of sparse data,an information entropy calculation method based on user score difference is proposed.The entropy value is used to compare the difference between users.(2)In order to improve the recommendation effect,this thesis proposes a Jaccard weighted similarity calculation method based on information entropy.This method takes into account the number of items jointly evaluated by users and the difference of scores between users.Through the traditional collaborative filtering recommendation algorithm Jaccard coefficient and based on The information entropy similarity algorithm performs linear weighting to achieve the recommended purpose.(3)Because the traditional collaborative filtering recommendation algorithm Jaccard coefficient only considers the number of users' common scores,and ignores the influence of user activity on project scores,this thesis proposes a Jaccard similarity algorithm based on user activity,and based on The information entropy similarity algorithm performs linear weighting to improve the recommendation accuracy.Finally,through the experiments on the public datasets such as MovieLens,it is verified whether the similarity calculation method proposed in this thesis is effective,and compared with several traditional collaborative filtering recommendation algorithms.The final result of the experiment verifies that the algorithm is effective,and the algorithm results are better combined with the common scoring project,user activity and user-to-user differences.