Research On WEB Load Test User Model Based On Data Mining

The performance of Web systems is very important.Web load testing is one of the necessary means to ensure the performance of Web systems.Therefore,Web load testing has important research value.The existing load testing methods have some problems,such as the user model is not real enough and the user model prediction method is lacking.For solving these problems,I proposes a user model for Web load testing based on data mining.Firstly,aiming at the problem that the traditional load test user model is not real enough,I propose a user model of interest feature based on Web log mining(UCIP-LTUM).The flow of the model is shown below: firstly,preprocess Web logs to remove useless data;secondly,statistic user behavior characteristics to prepare data for constructing user model;thirdly,construct user access sequence and build the basic framework of user model;finally,mine user interest access path to construct user model close to real users.Web load testing under the guidance of this user model can improve the accuracy of test results.Secondly,aiming at the lack of user model prediction methods,I propose a user model prediction algorithm based on Markov chain and Bayesian theorem(MAB-UPA).The flow chart of the algorithm is as follows: firstly,establish the correlation matrix of web page types to get the degree of correlation among web page types;secondly,use Markov chain to predict the type of web pages that users will visit;thirdly,use the Bayesian theorem to predict the specific web pages to be visited within the range of candidate web pages;finally,predict the user behavior characteristics of each page based on the existing user behavior characteristics data.The user model predicted by this algorithm is similar to the original user model.Finally,I experimented with the performance testing tool LoadRunner 11 on the Chinese numerical pool website.The user model is divided into UCIP-LTUM user feature model,MAB-UPA prediction model and irregular user model.And I compare and analyze the test results of each user model with the benchmark test results.The analysis results show that UCIP-LTUM user model and MAB-UPA prediction algorithm can effectively improve the authenticity of Web load test user behavior,and thus improve the accuracy of Web load test results.