| The collaborative filtering is one of the most widely used recommendation algorithms in current recommendation systems.It associates users with similar interests or items with similar characteristics based on the user’s historical interaction data,and uses these neighboring users or items to predict user preferences or behaviors.Although the collaborative filtering algorithm has achieved great success in the field of e-commerce,it still faces various problems,such as changes in user interest,information cocoon effect,and cold start.This paper mainly aims at the problem of the negative effect of information cocoon room caused by the change of user interest over time and fitting the change of user interest,and two models based on collaborative filtering algorithm are proposed.In the first model,in order to solve the problem that traditional collaborative filtering algorithms cannot calculate time-sensitive user interests,a collaborative filtering algorithm that integrates stable and changing interests of users is proposed.In the second model,the traditional collaborative filtering algorithm is unable to capture the user’s changing interests and ignores the negative effects of information cocoon rooms,and proposes a collaborative filtering algorithm for minority users’ interest changes.The key contents of this research are as follows:First of all,a collaborative filtering algorithm is proposed that integrates the stable and changing interests of users.The algorithm divides users’ interests into stable and changing interests,and models the users’ stable and changing interests respectively.The changing interest of the user is captured through the time window,and the interest of the neighbors within the time window is regarded as the changing interest of the user.The user’s stable interest is obtained through a probability implicit semantic algorithm that does not incorporate the time factor.Finally,the algorithm is based on the idea of weighted fusion,fusing the predicted scores of users’ stable interest and changing interest to generate the final prediction result.The comparative analysis of experiments on the datasets of Movie Lens and Netfix shows that the collaborative filtering algorithm that integrates the stable and changing interests of users can improve the recommendation accuracy of the model.Finally,a collaborative filtering algorithm oriented to changes in interest of minority users is proposed.While the algorithm calculates the user’s time-sensitive interests,it also avoids the negative effects of the information cocoon.The algorithm combines the three factors of time window,exponential decay function and user item interaction information with time effect to describe different users’ interest propensity values for different item types.Classify users through the improved K-means algorithm,and use a balanced method to increase the number of neighbors of such users and avoid the interest of minority users.In order to prevent the occurrence of information cocoon room phenomenon.The comparative analysis of experiments on the Movie Lens data set shows that the recommendation accuracy of the collaborative filtering algorithm for the change of interest of a minority of users is improved by 2.07% at the maximum. |
