Research Of Microblog User Interest Mining Based On Image-text Co-occurrence Data And Time Effect

With the rapid development of the Internet and social network,the speed of updating microblog information explosively increases.Users need to get content they are interested in through the platform.However,users face the problem of "information overload" because of the huge information flow.Based on such scenarios,it is an effective method for platform and users to analyze and mine user's interest tendencies so as to provide high-quality personalized information and commercial advertisement push services.In the microblog platform,users not only have long-term interest points,but also generate short-term new interest according to the changes of time and current popular things.With the development of the Internet,the status quo of multi-text short,more bars,more pictures and so on appears in the blog data.In the above environment,it makes good use of data characteristics to analyze and mine the current interest tendency of users,which has a good theoretical research significance.Firstly,this paper extracts the features of image and text data,and then proposes a multi-stage incomplete clustering algorithm based on Single-Pass(MIC-SP),which solves the problem of high time cost and order dependence of traditional Single-Pass algorithm.After that,the rule of user interest changing with time is fitted by function.Based on the idea of topic modeling,the time function is used to reduce the dimension of user topic probability distribution matrix.Finally,an algorithm using image-text co-occurrence data and time effect for microblog user interest mining(ICDTE-MUIM)is proposed,which makes full use of the image and text data generated by users,and calculates it according to the time effect of interest change.Probability of user interest tendency.Firstly,we design the framework of microblog data acquisition to obtain real data as experimental data set.Then,the MIC-SP clustering algorithm is compared with the traditional Single-Pass algorithm in terms of system overhead and clustering results.Then four evaluation indexes are set up,including prediction accuracy,missed detection rate,probability accuracy rate and topic difference.Four existing algorithms are selected as control group to verify the mining effect and performance of ICDTE-MUIM algorithm.Through comparative analysis of several groups of experiments,it is proved that the proposed mining method has more accurate positioning and more comprehensive coverage for microblog user interest mining,and shows better performance and efficiency.