Research On Double Similarity Recommendation Method Considering User Trust And Preference

With the advent of modernization of information such as hntemet+ and big data,the phenomenon of Data Overload(DO)has become an urgent problem to be solved.Recommended Technology(RT)is one of the current methods to effectively alleviate such phenomena.It can filter out non-target data from massive data and mine the useful value of Data Mining(DM)from potential information.The main research contents of the recommended technology include obtaining user records,matrix decomposition,similarity calculation,and prediction score.Among them,the similarity calculation is a key part of the recommendation technology:it compares the similarity degree of the user or the item,and pushes the recommendation list with high similarity to the target user,which affects the recommendation result.At the same time,tinder the prevalence of social networks,trust relationships are gradually integrated into the recommendation field.Therefore,considering user preferences and trust,it provides a new direction for improving recommendation accuracy.Based on the similarity calculation,this paper designs a dual similarity model that considers the user,s multi-source information,and proposes a recommendation method that can effectively improve the recolmendation accuracy.The main work of this paper is as follows:(1)Analyze the characteristics,Principles and applicable scenarios of existing similarity algorithms,and improve the similarity algorithm:Introduce the non-actual scores to be defined based on the modified cosine algorithm,add positive offsets,adjust the scores,and properly compress the overall algorithm;design the first type of screening protocol to filter out users with lower similarity values computed by the improved similarity algorithm.(2)Combining user preference and trust,a new dual similarity model is proposed:the improved similarity algorithm to calculate preference similarity model,and the trust similarity model is used for linear weighted fusion.Considering that the data is multidimensional and calculated There may be overlapping users,designing a second type of screening protocol to accommodate category data and exclude overlapping users.(3)Designing the dual similarity recommendation method of user trust and preference:In the preprocessing stage,the Lagrangian interpolation method is combined with the trust user to pre-fill the scoring matrix;In the similarity calculation,the trust and preference are built in the dual Similarity model;And,the traditional neighborhood scoring model is improved in the score prediction.Integrate the above to implement a new recommendation method.Experiments show that the new method has a good performance in terms of recomnendation accuracy and stability.