Research On Personalized Recommendation Of Rating Prediction Based On Information Fusion

With the continuous progress and development of mobile Internet,big data,cloud computing and Internet of things technology,the scale of information resources is expanding rapidly,and the problem of information overload is becoming more and more serious,resulting in the birth of personalized recommender systems(RSs).Collaborative filtering(CF)recommender algorithm does not need too much knowledge in specific fields,and it is easy to implement and has high accuracy,so it is widely used in the field of recommender systems.However,the traditional collaborative filtering has been facing the problems of data sparsity,cold start and scalability.It is difficult to effectively solve these problems only using the single rating information.With the continuous expansion of the scale of users and the continuous improvement of information technology,a large number of multivariate information has been derived,such as the demographic information of users,the type information of items,social network information,time context and so on.How to use the rich multi information to optimize the algorithm has become a research hotspot in the field of personalized recommender systems.In this thesis,based on missing ratings filling of sparse matrix,time context fusion,userbased collaborative filtering and item-based collaborative filtering recommender algorithms,we analyzed the shortcomings of existing algorithms and integrate multiple information into each stage of recommender algorithm.The specific research contents are as follows.(1)To solve the problem of data sparsity in recommender system,we proposed an effective pre-rating method based on users' preferences dichotomy strategy and average ratings fusion.First of all,we integrated multiple information such as users' rating information and the label information of items to construct user-preference matrix,analyzed and modeled users' interests and preferences.Then,we divided the items into two categories according to the parameterized dynamic threshold,namely user interested in and not interested in.For the missing ratings of these two types of items,different methods were used to fill in,which alleviates the negative impact of data sparsity to a certain extent.(2)The existing time-aware recommender algorithms ignore the difference of timesensitivity of different users.To solve this problem,we proposed a rating prediction method based on user time-sensitivity.Firstly,the time-sensitivity of users was analyzed and modeled by integrating multiple information such as users' rating information and the label information of items.Based on voting strategy,the cosine distance and relative entropy are used to establish judgment functions to detect the time-sensitivity of users.Then,the traditional collaborative filtering algorithm based on rating prediction was improved based on user time-sensitivity,and the combination parameters are optimized.Finally,we verified our methods on the standard dataset,and experimental results show that they could effectively reduce the prediction error and improve the quality of recommendation.