Microblog Recommendation Model Based On User Clustering With Multi-armed Bandit Algorithm

Microblog has become an important platform for people to share,disseminate and obtain information.The explosive growth of users has led to an exponential growth in the data generated by the microblogging platform,and the problem of information overload is becoming more and more serious.Therefore,in order to satisfy the needs of a large number of ordinary users,it is particularly important to provide accurate real-time recommendations for users.At present,most microblog recommendation algorithms are offline recommendations based on content or collaborative filtering,which does not make full use of user behavior information,resulting in prominent problems of user cold start and data sparsity.In fact,user behavior information is also a reflection of user interest.Fully mining user behavior information can make up for the lack of user historical data and alleviate the problem of data sparsity.Real-time recommendations can use user feedback to obtain user interest,and adjust the recommendation items according to the change of user interest,which is an effective way to deal with the problem of cold start of users.The user class is constructed,and the interest of the user class is used to represent the interest of ordinary users,which reduce the sparsity of the user data.A multi-arm gambling machine algorithm based on optimized exploration and utilization is proposed.Compared with the traditional algorithm,the accuracy of recommendation is improved to some extent.This paper proposes a method for generating microblog recommendation list,which generates microblog recommendation list for current users,and updates the historical feedback information based on the user's feedback to the recommendation list.The main work of this paper includes three aspects:(1)A complete user class is constructed based on Page Rank algorithm and K-means algorithm.Firstly,an important user discovery algorithm based on Page Rank algorithm is proposed to find important users in the microblog network.Then,an improved K-means algorithm is proposed to cluster important users and construct important user classes.Finally,the similarity between ordinary users and important users is calculated according to users' concerns and personal attributes in Weibo,and ordinary users with higher similarity are added into the important user class to build a complete user class.(2)Microblog Recommendation Model Based on Multi-arm Gambling Machine Algorithm.Firstly,a multi-arm gambling machine algorithm is proposed to optimize exploration and utilization.Then,based on the algorithm and user clustering,a method of generating microblog recommendation list is proposed to provide personalized microblog recommendation for users.Finally,the recommendation model and its improvement model are constructed by integrating user clustering and multi-arm gambling machine algorithms.(3)Sina weibo real data set is captured for accuracy and validity verification.We set five comparison tests for microblog recommendation:(1)A microblog recommendation model based on user clustering and multi-arm gambling machine algorithm and an improved microblog recommendation model based on user clustering and multi-arm gambling machine algorithm.(2)An improved microblog recommendation model based on user clustering and multi-arm gambling machine algorithm and four random exploration algorithms.(3)An improved microblog recommendation model and three confidence interval algorithms based on user clustering and three confidence interval algorithms.(4)An improved microblog recommendation model and three confidence interval algorithms based on user clustering and three probabilistic matching algorithms.Experimental results show that the improved model can recommend blog posts that users are interested in,and the recommendation effect is5.62%,5.43% and 33.37% higher than the existing random exploration algorithm,confidence interval algorithm and probability matching algorithm,respectively.