Modeling The Evolution Of Traffic Data On Visitor Behavior

Among diversified business application of the internet, shopping on the internet provides the most indirect contact between customers and companies. With the characteristic of high interchange ability, network enables operators of networks to find out the actual visiting process and the periods of customers through net data. Meanwhile, marketing personnel can use theses collected data to forecast purchasing behavior of the customers. With in-depth analysis of the customers' behavior, marketing personnel then may forecast the royalty of the customers so that they can do their best to retain the clients. Based on such background, this thesis analyses click stream of some certain websites, and establishes a probability modal of evolving behavior. This modal makes an analysis of visiting behavior of those who visit the websites, and then forecasts their mutative process along with time.Firstly, this thesis generalizes the related research from home and abroad, and then summarizes advantage and disadvantage of each method. By the reference of these points, the problems that will be solved in this paper are put forward. Secondly, the thesis describes the conception and characteristic of internet marketing and database marketing based on network. Then comes the analysis of data structure of clickstream. On the basis of all the above, this thesis brings about the mathematics description of the statistical method. Exponential distribution, gamma distribution and maximized likelihood estimation (MLE) are chosen to be the main components of the statistical distribution modal. The modal is established from two parts that are heterogeneity of customers and dynamic time and schedule. Then comes the mathematic analysis of evolving visiting behavior. Through MEL, the thesis estimates the static parameters of the modal, then analyses the likelihood function and estimation which is added dynamic multiplier. At the end of the thesis, the demonstration is made for the evolving model. The data in this thesis is about several-month visitor traffic from two representative business webs. Above all, the analysis is based on the static exponent-gamma distribution, and then comes the conclusion that the static model can't capture the heterogeneity in visiting process. And then, the thesis compares evolving model with other different model, including static exponent-gamma distribution andconsidering customer exiting. In this thesis, the modeling efficiency is validated by the extended forecast ability for the evolving visit behavior. The example proves the fact that evolving visit model is better than static exponent-gamma distribution model in the forecasting aspect.Finally, this thesis gives the conclusion that the behavior of visitors varies with different time and experience. By the use of parameter estimation, the thesis indicates that customers' visiting rate is gradually decreasing during a definite period, in which the entry of new visitors conceals the leaving of many old customers. All the above makes an important sense for the managers of e-business websites, consequently they can realize how to keep clients in the process of internet marketing.