The Key Problems Research On Expert Finding In Microblog

Recently, with the development of Web2.0, social network became more and more popular. Microblog such as Twitter and Weibo was widely used. Social medias change people’s life style and produces huge social influence. There are millions of users in social network, including celebrities, sports-person and organizations. How to infer experts with a given topic is a hard and challenging problem.In this thesis, we focus on Sina Weibo platform, finding experts through users’ tags and follow relationships. The investigated data shows that users in social network tend to follow those who share some common interests with them. And we give the assuption that users in Weibo are homophily.Sina Weibo platform allows users to tag themselves and for a weiboer, these tags are usually highly conclusion about him. His interest can be reflected on these tags. We improve the TF-IDF model to filter the tags with low equality. Under the homophily assumption, we design a tag propagation algorithm to expand users’ interest tags, mining users’ interest graph. Then it is possible to infer expert through potential semantic information about the follow relationship.With respect to Pareto principle in social network, we set users into kernel communities and auxiliary communities. Combined with interest graph, we raise kernel community based SALSA algorithm to compute user’s influence to a given topic. Finally, we propose a model containing some features to do the finement.We crawled 40 million Weibo users’ information, including profile and follow relationships. And we manually label a set of words to do the test. The result shows that our models are efficient. At the same time, our models can scale well with more users.