Content-Based User Profiling

As a result of the great evolution in the information technology and computing fields in recent years, people began to step into an age of overload-information. Both the information consumer and information producer are suffering the bad effects of this. For one side, the users may find it becomes more and more difficult to pick out what they want and enjoy as the Internet is full of the useless information; and the other side, the information producers can hardly let their information appears to the target users. The personalized recommender service is born at the right moment.While a good personalized recommender system should have a good knowledge of every target user, it is important to profile the user’s interest correctly. What’s more, the precision and diversity of the recommender system extremely depends on whether it has a good user profile. However, it is not an easy work to have a good user preference model for the constraints of the input data.In this thesis, some research work has been done on the traditional modeling algorithms which extracts user features from the support content. Firstly, we recomplete the algorithm on the Sina Weibo dataset and Douban Movie dataset, then we combine the related theory of the word embedding to our work. The comparison proves that our work can be successfully applied into recommender system.Our research is expending research area of the user preference modeling, and, to some degree, inspire the following work.