Research Of Personalized Microblog Recommendation Technique Based On User Analysis

As a typical representative of the social network, microblog has attracted a large number of users by it’s facilitate communication, spread rapidly and interactivity. With the growth of user scale, information in microblog also shows explosive growth. Large amount of information leads to the information overload in microblog. Information overload brings many problems to users, one of them is that with t he number of friends that user followed increased, the daily information pushed to users is also increased. Microblog users are overwhelmed with mostly uninteresting pieces of text in order to access information of value. How to pick out the contents that users are interested from a large number of microblog information is the key to solve the information overload in microblog.To provide the personalized recommendation service for microblog users, firstly, we must know why users show great interest on this microblog information. The reasons maybe are the relationship between users, the influence of information publisher or the similarity of interest between the user and the topic of microblog information. In this thesis, we convert the microblog recommendation into a sort problem on the information stream that user received. The learn to rank method is used to learn the sort function on training data, then we use the function to calculate the relevance score that representative the relevance between microblo g information and user preference. The recommendation list is given by a descending sequence of microblog information in user session. The research contents of this paper are as follows:(1) In order to quantify the user ’s characteristics that can embody user ’s personalized demand, we analysis the existing research methods and then select proper feature representation methods.(2) We study on the method to partition user ’s session. Since our purpose is to recommend the users recently received information, we have to find out what information the user receives in the recent visit. We use an approximate method to define a user session based on users’ observable activities such as publishing, replying.(3) The relative relevance assumption is introduced to construct the training data. By analyzing users’ action in microblog, we use the relative relevance assumption to label the microblog data. Under the assumption, the microblog date is represented in pairs with partial ordering relation.(4) Learn to rank method is used to solve the multi- feature fusion problem. On the training data, we use learn to rank method to get the relevancy ranking model. Then we use the model to calculate the relevance score of each piece of microblog information in user ’s session, and finally we give a descending sequence according to their scores as the recommendation results. The experimental results il ustrate the effectiveness of our method.