The Design Of Information Push System Based On WiFi User Behavior

With the rapid development of computer technology, especially the network technology, people’s living conditions and social environment have been changed enormously. Wireless network spreads rapidly in every aspect of people’s lives and has an increasing number of users because it possesses the features of high speed, high reliability and no need of paying to the telecom operators for the data. At the same time, the number of websites has been increasing and the content of these websites become more and more complex. The Internet users on the one hand have difficulties in finding information they really need or are interested conveniently and efficiently, on the other hand they may have to pay more money for the cost of Internet traffic. "Abundant data, insufficient knowledge and lack of pertinence" has become a very serious problem. Aim at the current problem, a kind of information push service which is based on the analysis of users’ network behavior has come into being. It provides a brand new service mode for the users which can extract and analyze potential valuable information from vast records of users’ online behaviors and solve the problem of information overloading.Through the study of the use of WiFi, current push service and the demand of users, this thesis tries to combine the WiFi technology with information recommendation system and set up an information push mode which is led by both features of local area network (LAN) and the analysis of users’ network behavior. First, separate the system concerns to WiFi gateway and server. The server’s function is to grab the data of Internet users, manage the user access data and resource information and implement recommendation algorithm. And then in the process of modeling based on collaborative filtering recommendation algorithm, due to the similarity of the time forgetting curve and the change of users’ interest over time, this thesis applies the time forgetting curve to implicit rating calculation process, which considers the changes of users’ interest over time on the basis of the traditional information recommendation algorithm. Under the circumstances of less information, this thesis uses the methods of "set fixed default values" and "mode number method" to solve the problem of recommend model of sparse matrix and cold start.In the process of searching nearest neighbors set, through improving the cosine similarity algorithm promote the accuracy of push results.In the end, use TcpDump to grab the users’ network behavior data, parse the packet, establish a "user-item" matrix, build the user model, edit code for the recommendation algorithm and get the final recommended results. By using the information push system which is based on WiFi users’ network access to do the predict training and analyzing the results, a conclusion can be drew that the improved collaborative filtering recommendation algorithm has better prediction effect.