Research On Micro-blog Recommendation Method Based On Topic Model

As the most typical social network service media in the Web 2.0 era,micro-blog is integrated into people's life.Users can disseminate and get information at any time and anywhere,it is an important way to achieve social interaction.As micro-blog growing rapidly,it contains abundant and valuable information,how to achieve micro-blog recommendation for users from these huge amounts of information has become an important research problem.This paper is the study based on this kind of background.The main work and contribution of this paper are as follows:(1)Based on the study of related topic models and their applications of recommendation,a micro-blog recommendation method based on RPMPS model with LDA topic model and KL divergence is proposed.In order to ensure the real-time performance of the method,the micro-blog is filtered in the process of data processing to improve the response time.The user's interest set is defined,tags of micro-blog publisher acts initial classification of micro-blog.The similarity between user information and the topic of candidate micro-blog is obtained from the document-topic probability distribution matrix,and the similarity between user information and the contents of candidate micro-blog is got by computing word frequency probability through the document-word matrix.Finally,the recommendation is completed by ranking the overall similarity between user's current interest and candidate micro-blog constructed by topic similarity and content similarity.(2)Because micro-blog has the characteristics of short text,the performance is easily affected by micro-blog sparse feature when modeling them.In order to further improve the performance of micro-blog recommendation,the BTM topic model is studied,and a novel clustering algorithm is introduced,then a micro-blog recommendation method based on CFSFDP clustering and BTM topic model is proposed.The idea of trust degree between users is constructed,the recommendation results is effectively improved by adding trust degree between users.Firstly,CFSFDP clustering algorithm is used to clustering the micro-blogs to select the micro-blog group which is similar to the user's interest to form candidate set.The topic similarity and the content similarity between candidate micro-blog and user's current interest are calculated through BTM topic model and KL divergence.And then the overall similarity between candidate micro-blog and user's current interest with user's trust degree.Finally,micro-blog recommendation list is formed through ranked similarity.(3)The experimental data is crawled from Sina micro-blog,to verify the proposed micro-blog recommendation method based on RPMPS and micro-blog recommendation method based on CFSFDP clustering and BTM topic model respectively.The experimental results show that the two micro-blog recommendation methods can obtain good recommendation performance.