| The accurate classification of users is very important to improve the quality of customized services.However,because of privacy protection,users often do not cooperate with network service providers and refuse to provide personal information such as location information and hobbies.To solve this problem,this thesis makes the following work:First of all,this thesis deals with user information in the wireless Internet,and analyzes user’s information of HTTP with the promise of users’ privacy protection.It extracts user’s IMEI,Userid and Appid information to set up the user fingerprint collection device,which can determine the IP address of the user is used.Then it extracts the characteristics of the network layer by IP address as a user and extracts the analysis of user’s application layer traffic information.It uses the random forest algorithm by using network layer statistical information to predict the user’s location type(apartment,campus,etc.).At the same time,it also uses K-means algorithm to cluster users into eight types with the keyword classification information of user’s application layer URL.And it can calibrate the user type and ultimately determine user types with each URL keyword categories percentage range in the eight types.Finally,this thesis analyses the advantages and disadvantages of user interest determination through the user types and location type.In order to improve the prediction accuracy for the user,this thesis summarizes location types and user types effect of user preferences after analyzing two aspects of commonality and diversity of user interest determination.Experiments show that this scheme can adapt the user type and geographic location type.It can also improve the accuracy of user behavior analysis by correlating their type and geographic location. |
