The Design Of The User Recommendation System Based On Network Users Behaviors Analysis

Nowadays, the rapid development of IT and Internet technology has provided people with a myriad of opportunities and wealth, but at the same time, it produces billions of data every moment. If there is not a good method to explore the potential information therein, it will not be able to provide better service for users. It is a good way to extract and analyze interaction behavior data between specific Internet user and the Internet from the massive data, and It also can be an effective means to study and explore the user’s interest and demand; according to the specific user’s interests and demand, recommendation system can provide users with the "personalized" service by the Internet platform. This paper research includes:(1) The recommendation system based on content is researched. First of all, the.Pcap file and packet structure are studied. After the studying of HTTP and TCP protocol communication principle, we extract some key data like the webpage text URLs. Then we analyzes the relative algorithm of the feature extraction and feature dimension reduction, get the feature vector which consist of feature words and corresponding weights, and construct the Vector Space Model(VSM). Next, we research the relative theories and the available algorithms of the text clustering, and the first text categorization model is established by combining with the VSM, the BWP index, the k-means algorithm, the cosine measure, the square error criterion, etc. In the recommended links, the recommendation system based on content is studied, and the recommended method is determined, then we complete the design of the user recommendation system based on network users behaviors analysis.(2) The secondary clustering recommendation algorithm based on information entropy is proposed. Through the researching on the relative theories of information entropy, the fact that information entropy can be regard as a kind of measure standard and quantitative measurement of the accurate recommendation is cleared, and thus the theoretical foundation for the secondary clustering recommendation model is established. Then we propose and use the relative concepts and formulas of information entropy, incloding the nearest entropy difference, the threshold value of the nearest entropy difference, the average entropy value approximation, and so on, to judge and calculate the initial clustering cluster numbers and hearts and the final recommended contents. And then the second text categorization model is established by combining with the VSM, the threshold value of the nearest entropy difference, the average entropy value approximation, the continuous random variable of uniform distribution, the k-means algorithm that have already determinated the initial clustering cluster numbers and clusters hearts, etc. Next we get the number and contents of the recommendation results by fitting the logarithmic function and combining with distance and information entropy measurement. Finally, the structure of the secondary clustering recommendation model based on information entropy is completed.(3) Some relative experiments is designed to verify the performance of the user recommendation system based on network users behaviors analysis and the secondary clustering recommendation algorithm based on information entropy. The experimental results show that the user recommendation system based on network users behaviors analysis designed in this paper can implement relative recommendations for specific user successfully, and provide recommendation results which are highly similar with the specific user’s interest and demand. By comparing the precision, recall rate and F-measure coefficient of the secondary clustering recommendation algorithm based on information entropy with the traditional algorithm, we find that the former have more obvious advantages in these three parameters, which prove that comparing with the traditional algorithm, the secondary clustering recommendation algorithm based on information entropy improve the recommendation accuracy.