Users' Interest Detecting And Friends Recommendation Research Based On Topic Model

Recent years have witnessed the fast development of Internet technology along with the development of science and technology.Social network media is expanding rapidly and it has become an integral part of life.People want to meet more people and discover more fun in social networking.Hence,it is a key task to recommend new friends to users based on their characteristics.And it's also users potential needs.There are three ways for users to make new friends in present social networking.On the one hand,users can add new friends manually.On the other hand,social networking site can offer recommendation according to the site's top events or users' characteristics based on their recent updates.However,this recommendation algorithm fails to take into consideration users' unique individual characteristics.It is a key direction of the social network research to offer personalized recommendation based on users' characteristics,which can significantly improve the users' experience.And the success of NetEase music sets a good example.It obtains a large number of users by finding users' unique musical DNA.In recent years,user interest model arises in the field of friend recommendation.It classifies users via their interests,and it prevails over the traditional Content-based Recommendation.This thesis puts up with an algorithm on new users' interests detecting and interest mining based on previous researches.This thesis starts from users' interest distribution as well as the arising and changing of their interest.And the experiments prove the serviceability.The main contributions of this thesis are as follows:1.Firstly,this thesis presents an improved LDA topic model that can be used in new users' interest distribution.2.Secondly,this thesis presents an improved algorithm which functions well in discovering users' potential interests and the emerging interests.3.Thirdly,this thesis designs a tool which can grab and handle micro-blog's data.This tool takes advantage of the existing structure of Sina micro-blog's page and the existing natural language processing technology,and it is the basis for the next experiment.4.Lastly,these improved algorithms have been proven to be very useful in predicting users' interests and exploring their distribution by the simulation experiment with the experimental data.