Network User Behavior Analysis Base On Traffic Monitoring And Measurement

With the development of network technologies and services, the number of network users increases year by year. People can not only browse news, listen to music, watch videos, but also deliver their comments, purchase goods and play games. The Internet has become an extension of real life. People interact in the online world and show a variety of network behaviors. The study of network user behavior has great significance for network optimization, network service design and network service advertising.Traffic monitoring and measurement technology can provide data reflecting characteristics of network traffic. This technology was first applied to network planning and QoS measurement by Internet Service Provider (ISP). In recent years, the emphasis of ISP are transferring from network building, operating and maintaining to provide high quality services for large group of network users. In this condition, network user behavior analysis is particularly important for ISP. Meanwhile traffic monitoring and measurement can obtain real and reliable data for network user behavior analysis.The thesis presents the research on the network user behavior analysis. The data support our work is collected from MAN environment and our conclusions can reflect the current situations of network behavior in China.The main contributions of our work are as follows:Statistical analysis of network user session behavior:Analyzing network user session behavior helps to understand people's basic usage of network, while the current relative research is still inadequate. To solve this problem, we analyzed MAN network user session behaviors with the distribution of session indicators over time, correlation between session indicators and Pareto effects in session behaviors using statistic methods. Our conclusions are of great value to recognize user session behaviors in China.Analysis on network user time span preference:Lots of works focus on studying network user behaviors in different time spans, but there is lack of time usage mining on network. Our work revealed and verified the close-similar property of network user on-line time distribution based on the analysis of users' time usage data, and developed Hierarchical Clustering based on Fast Grouping algorithm (HCFG). According to the experimental environment, we proposed Time Span Coincide Ratio to evaluate if each cluster had significant time span preference. Also we evaluated HCFG with other general criterions such as Cohesion, Separation and Time complexity. The evaluation results show that HCFG can distinguish preference time span in different clusters and is efficient enough to discovery network user time span preference patterns.Analysis on network user Web preference:ISPs and Websites can improve the performance of service customizing and direct marketing by understanding network user Web access preference. While most of the works aim to a single Website, this work analyzes network user Web access preference to multiple Websites. Within the statistical analysis on Web usage data, we ranked the popularity of Web types and found that most users access to a small scale of Web types. Clustering is an important method to detect the mainly patterns in network user Web access preference. Yet data compression is a problem when facing high-dimensional data. We introduced Closed Frequent Itemset to solve this problem and developed Closed Frequent Itemsets Hierarchical Clustering based on Quantities algorithm (CFIHCQ). This algorithm can not only compress the data dimension but also discovery stable user behaviors. The result generated by this algorithm can reflect each group user Web type preference intuitively. We evaluated the algorithm comprehensively including:cluster result, parameters and complexity. The evaluation indicates that CFIHCQ has many advantages such as predicating user Web access behavior accurately, constructing clusters compactly, showing user Web preference significantly, choosing cluster number flexible, and mining large data efficiently. Analysis on network user Web access state transition pattern: Dynamic behavior analysis is a hot point in network user behavior analysis. Lots of work study user navigation pattern on certain Website, but there are not enough research on user dynamic behavior in the whole internet. Our work is the first time to analyze network user Web access state transition pattern on the point of the whole internet. We did statistical analysis on Web access state sequence of continuous on-line users, and found that users' Web access state can be predicated by history data. Then we purposed the concept of Continues Web-access State Transition Matrix (CWSTM), and developed CWSTM clustering algorithm based on entropy. This algorithm improves the identification number of deterministic transition probability by introducing entropy to CWSTM. Experiment result shows us that the more deterministic the transition probability is, the better predicate results are, which improves that our work is very significant.