Research On Diversity In Recommendation Algorithms

Recommendation technology has become a key technology for providing personalized services in the age of information overload.The current recommendation system pursues high-accuracy recommendation,which leads to a decrease in the diversity of recommendation results and a poor user experience.In view of the low diversity of recommendation results in the current research,this paper applies the social curiosity theory to the recommendation of similar users in individual recommendation,and proposes a diversity recommendation algorithm based on the curiosity of similar users(Similar users' curiosity).In group recommendation,combining re-ranking method based on item popularity,a group recommendation diversity method based on item popularity is proposed(Group recommendation method based on item popularity).Based on the above two aspects of research,the main work of this paper is as follows:(1)Aiming at problems that are difficult to obtain and relatively sparse among users,such as friends and trust relationships,propose a diversity recommendation method based on the curiosity of similar users(SUC).Based on similar users and collaborative filtering methods,SUC method calculates the user's projected prediction score and curiosity score,and then integrates the user's project diversity score and project recommendation list.SUC method calculates the similarity relationship between users by analyzing the user's rating data,without additional data,and enhances the universality of the method.Experiments on five real data sets show that compared with the benchmark method,SUC method improves the diversity of recommendation results,and also improves the accuracy of recommendation,recall and coverage.(2)Aiming at the problem of increasing the accuracy of group recommendation while ignoring the diversity of group recommendation results in group recommendation,a group recommendation diversity method(GRMP)based on item popularity is proposed.First,the GRMP method uses the traditional recommendation prediction method to obtain the user's prediction score for the item.Then,predict the predicted score of the item by the prediction group.Finally,,the group recommendation list is regenerated according to the reordering method based on the popularity of the items.Experiments on two real data sets show that compared with other benchmark methods,GRMP improves the diversity of recommended results while ensuring minimal loss of accuracy.This paper proposes two diversified recommendation methods.It is verified through experiments that both SUC and GRMP methods can improve the diversity of recommendation results,and the SUC method can also improve other recommendation indicators.