Research On Recommendation Algorithm Based On Project Feature Fuzziness And User Interest Fuzziness

With the rapid development of the Internet,the problem of "information overload" is becoming more and more serious.As an important tool to solve the problem of information overload,recommendation system will play an engine role in the development of the Internet in the future.The design of recommendation algorithm is the most important part of recommendation system.At present,the most widely used collaborative filtering recommendation algorithm has achieved remarkable results in the field of recommendation system.However,there are still some shortcomings when facing the problems of data sparseness,the transfer of user interest over time,the vagueness of project characteristics and the vagueness of user interest.This paper proposes a recommendation algorithm(CFIUM)based on the fuzziness of item features and user preference.The specific research content mainly includes the following aspects :(1)In view of the sparsity of user item score matrix and the migration of user preference with time factor,this paper adopts the improved time factor attenuation model based on Newton cooling law and SVD(singular value decomposition)technology to carry out the score weighting and filling of user item score matrix.The similarity of scoring information between items is calculated according to the filled user item scoring matrix.(2)For the fuzziness of item feature information,gaussian-like membership function is used to construct item feature membership matrix,and the similarity between item feature information is calculated according to item feature membership matrix.The similarity of user rating information and project feature information is weighted by the optimization algorithm of grid search to obtain the com prehensive similarity between the two kinds of information.(3)According to the user's historical behavior information,the user item feature membership degree is used to construct the user item feature preference matrix,and use the project characteristics of the user preferences matrix of weighted after filling the user program rating score matrix correction.Based on the trapezoidal membership function,the user interest model is constructed based on the revised score data.Finally,the recommendation trust score calculation strategy,which comprehensively considers the similarity between the user's interest model and the project,is used for recommendation.(4)Based on the above theoretical methods,the proposed algorithm was simulated on Movie Lens100 k movie data set,and its recommendation performance was evaluated through experimental comparative analysis.Firstly,the influence of user interest variation parameter and similarity weight coefficient on recommendation effect was analyzed by comparative experiment,and its optimization parameters were established.Then,The algorithm in this paper is compared with the project-based system filtering algorithm(ICF)algorithm and the recommendation algorithm(FIUM)algorithm based on the fuzziness of project characteristics and user interest,respectively,under different number of training sets and different top-N recommendation numbers.and the experiment shows that the recommendation algorithm has improved the recommendation accuracy and recall rate compared with the comparison algorithm.Compared with ICF algorithm and FIUM algorithm,the average accuracy of CFIUM algorithm is improved by 11.49% and 7.77%respectively.The average recall rate increased by 5.29% and 4.99%,respectively.The main innovation points of this paper are as follows: A recommendation algorithm considering the fuzziness of project characteristics and user interests is proposed,and the similarity between projects is calculated by integrating user rating information and project feature information through the optimization algorithm of Grid search.