Research On Personalized Recommendation Method Based On Diversity

With the rise and rapid development of Internet technologies,including cloud computing and big data,data and information on the Internet have exploded.There is an inestimable value hidden in Internet information data,which provides all-round convenience for people in this era,but it also causes serious “information overload”.At this time,the recommendation system came into being and has been widely used,which help users discover new items of interest to them.Accuracy-based recommendation is the most widely used recommendation method,but relying solely on accuracy may cause overfitting of recommendation results,thereby reducing users' satisfaction.As a result,another evaluation index of recommendation methods,recommendation diversity,has been paid more and more attention by researchers and network platforms,and plays an important role.The improvement of recommendation diversity can allow users to obtain a better personalized experience,increase user loyalty to the website platform,help users discover more long-tail products of interest,and improve the overall profit of merchants.However,the diversity-based recommendation application is becoming more and more extensive and there are also problems such as ignoring the influence of time factors,incomplete calculation of user similarity,and local optimization of user clustering.In order to correspondingly solve the above problems,this paper proposes personalized recommendation method based on diversity,so that the recommended results can not only ensure that the accuracy does not decrease,but also can improve the aggregate recommendation diversity.The main innovative researches in this paper are as follows:(1)Aiming at the problem that users' interest in items migrates over time,an improved logistic weight function is introduced as a time factor to correct user ratings,so that the proposed method can increase the recent user-item rating weight as much as possible and reduce long-term ratings.The user-item ratings weight accurately reflects the user's current interest and preference characteristics and improves the accuracy of the recommendation results.(2)Aiming at the problem of incomplete user similarity calculation in collaborative filtering recommendation,the user-category preference matrix is ??calculated according to the user's preference for items,and the similarity of user ratings and user category user preferences are reasonably integrated,so as to obtain the comprehensive similarity of users overcomes the shortcomings of traditional collaborative filtering recommendation that only focus on user ratings and ignores users' interests and preferences,and improves the comprehensiveness and accuracy of user similarity calculation.(3)Aiming at the problem that the clustering results are easy to fall into the local optimum when using the spectral clustering algorithm in the recommendation process,the adaptive local scale parameter is used to alleviate the sensitivity of the scale parameter,and the weight factor between users is introduced to suppress the influence of outliers.Spectral clustering can improve the accuracy of user clustering,increase the discrimination of nearest neighbors across classes,and achieve the goal of improving the diversity of target users' neighbors and thus the diversity of recommendation results.Experiments were carried out on the Movie Lens data set,and the optimal parameters of the algorithms based on comprehensive similarity and spectral clustering optimization were determined,and then the method proposed in this paper was compared with the other four recommendation algorithms in terms of accuracy and diversity.Analysis and experimental results show that this method can improve the aggregate diversity of recommendations on the basis of ensuring accuracy.