Research On Personalized Friend Recommendation Algorithm And System Implementation Based On Micro-blog Data

With the popularity of micro-blog,We-Chat and other social networks,everyone has the chance to become a publisher and disseminator of information,and these social services thus accumulated a large number of user-generated data,including user profile,social relations,text content and so on.Such a large number of user data has become a significant source of information for industry,they can use these data for large data analysis,so as to understand the user,but also provide a lot of data to support academic research.But a fact cannot be ignored is that even though social networks has brought convenience to people's lives,at the same time however,the overload problem of network information has become increasingly serious.Faced with a flood of information from which it is hard to find what you are really interested in,personalized recommendation system has been used as an effective means of solving the problem in all kinds of social networking sites.In this paper,in order to solve the problem of the sharp increase in the number of micro-blog users,we use the content of micro-blog and the link information to recommend potential friends to users.The details and conclusions are as follows:(1)Research on the friends recommendation method based on the users'blog topic and the information of network structure.Due to the arbitrariness and falsity when people describe the attributes of users,it is invalid for us to apply the node attributes effectively while building the model of user interests.Aimed at this problem,a sort of hybrid friend recommendation method was proposed based on the interest preference and structural closures.At first,by using the LDA topic model,a model was constructed for users' micro-blogs,so as to mining the interest of users.At the same time,the method refined the preference of the target users according to the principle of homogeneity through the interest preference of friends.Meanwhile,a novel prediction index based on network structure was proposed to measure the structure closeness between users.The experiment results indicate that the rate of accuracy and AUC has significant improvement considering the interest of user,compares with the effect of recommendation via only on the partial structure.The interest preference of friends has also to some extent enhanced the interest mining effect of part of users whose blog topics are not explicit.(2)User interest mining method based on improved Link-LDA topic model.LDA topic model only uses the text content to dig the user's interest distribution but ignores the friends-following information that already exists.Aiming at this problem,the paper gives up the traditional LDA and draws on a novel Link-LDA model,and makes some improvements to make it more adaptable to the mining of user interest in social networks.The experimental results showed that this novel theme mining algorithm can more effectively describe user preferences on different topics,and obtain higher recommendation accuracy in the friend recommendation task.(3)The design and implementation of micro-blog friend recommendation system.The paper uses B/S structure,spring-MVC framework to design a micro-blog friend recommended system.The system simulates the real social network,recommends new friends for user according to the algorithm posted in this paper,and shows the results and the related recommendation interpretation in the recommendation module.