Research On Personalized Recommendation Algorithm Based On Time And Location

With the progress of society and the growing variety and quantity of information,it is becoming increasingly difficult for people to obtain information they are interested in from the large volume of data.The application of recommendation algorithms has alleviated this issue and has been recognized and approved by the industry.However,there is a sharp rise in the number of problems with the in-depth study of the algorithms.The impacts of both time and space factor on recommendation precision are commonly ignored by traditional algorithms.It is found that the interests of users from different regions might vary a lot,and in addition,their interests usually evolve with the time,which leads to the issue of lower recommendation accuracy.And besides,the trust relationships can also affect to some degree the performance of recommendation systems,since many users tend to ask their friends for advice and recommendations.To the above problems,this paper focuses on the impact of time,geographical location and trust relationships on personalized recommendation algorithms.The contributions are divided into three parts: Firstly,the recommendation algorithm integrating time based latent semantic completion model with subgroup partitioning.The prediction matrix is obtained by the filling of the missing entries of the original rating matrix on the basis of the implicit semantic model fusing the time factor.Then the matrix is partitioned into subgroups by the chosen clustering algorithms and the recommendation list is generated by collaborative filtering algorithms accordingly.The second work is about the research on personalized recommendation algorithm integrating the factors of time and position.Based on the prediction matrix of the previous work,the recommendation results are further filtered out by the collaborative algorithm which takes the user's position into consideration.Thirdly,the exploration of trust relationship based recommendation algorithm.The trust relationship between users,as well as the implicit trust relationship between users' rating are fully exploited and considered to further improve recommendation precision based on the gSVD + + algorithm.The explored problems and the corresponding solutions of this paper are summarized as follows:1?To the problems of data sparseness and lower recommendation accuracy in traditional personalized recommendation algorithms,the missing entries of the original user-item rating matrix are predicted and filled applying the proposed implicit semantic model integrating the time factor.The process effectively alleviates the problem of data sparseness,and at the same time,the fusion of the time factor effectively depicts the variations of user preferences over time.Then objects with similar preferences and interest characteristics in the predicted rating matrix are partitioned to the same subgroup by the selected clustering algorithms.The recommendation lists are acquired for target users by collaborative filtering algorithms in the subgroups.2?In view of the problem that temporal and spatial factors can affect the accuracy of recommendation algorithms,the model takes a comprehensive consideration of the influence of time and location transformation on user preferences.And a personalized recommendation algorithm is therefore put forward which fuses the temporal and spatial information.In view of the prediction matrix obtained by the proposed implicit sematic model fusing the time factor,the users' locations are determined according to their postcode information.Specifically,users are allocated to the nodes in each layer of the pyramid by the pyramid model,and local recommendations are made by the collaborative filtering algorithm for each node,and finally,a comprehensive recommendation is executed to acquire the final recommendation list by the assignment of a weight to the recommendation results of each layer in the pyramid model.This algorithm has obvious advantages in accuracy and can improve the satisfaction of users.3?To the problem that trust relationship between users can improve the recommendation accuracy,we put forward a personalized recommendation algorithm fusing the trust relationships.The proposed algorithm fully takes into account the characteristics of explicit trust relationship,implicit trust relationship,and trust propagation among users on the basis of matrix factorization.The trust relationship between users and their implicit trust relationship by the mining of their ratings are combined into gSVD + + by the weight factors.The experimental results show a significant increase in the accuracy of the algorithm.In brief,the innovations of this paper are as follows:1?The effect of time factor is analyzed in detail and integrated into the implicit semantic model.Objects with similar preferences or interest characteristics are partitioned into the same subgroups by clustering algorithms,which alleviates the issue of data sparseness and improves the recommendation accuracy of the algorithm.2?The time and geographic location are combined according to the dynamics of user preferences,and a personalized recommendation algorithm based on time and geographic location is proposed to meet the needs of temporal and spatial variations of user preferences.3?By the fully mining of the trust relationship in social networks,the implicit trust relationship between users and their implicit trust relationship obtained by mining their rating information are combined into the matrix factorization algorithm,which effectively improve the recommendation accuracy.In this paper,the effects of various factors on the accuracy of recommendation algorithms are discussed in detail based on the above research contents and contributions.A recommendation model integrating time,location and trust relationship is built.The experimental analysis shows that the research in this paper can further improve the accuracy of recommendation systems.