Research On Method Of The Web Access Object Trajectory Clustering

Since the birth of the Internet more than20years, Web-based information systems,E-commerce and Web Services have gotten the rapid development. At the same time, theInternet has hoarded vast amounts of the click-stream data and user data because of the hopsadvances in data collection and data storage technology. That’s a huge challenge to extractvaluable patterns from tens of trillions Web data. The present study is divided into three maindirections: the Web-based structure mining technology, contents mining technology and Webusage mining technology. And we focus on the last one.To initiate the study, the main clue to the paper lies on the Web user access datapreprocessing, access trajectory clustering technology and access personalizedrecommendation. The main research results are elaborated from the following aspects.Firstly, after in-depth study to the theory and technology of the web usage mining, wedesigned the overall framework of the mining implementation process, including datapreprocessing, trajectory clustering process, and personalized recommendation core module.During the first stage, we achieve the data formatter, data-element identification, dataintegrity sub-processes and the access transaction identification, etc. algorithms, andaccompanied by a text icon elaborated. Based on the public server log datasets, we conductedseveral experiments to verify the correctness of the proposed program and get the finegranularity of access transaction sets, which lays a solid foundation for the late work.Secondly, to face the problems such as the huge data, the clustering process not efficientand taken up a lot of system resources and the assessment method to the accuracy of theclustering results, we propose a new formal way to represent user interest feature andcalculate user similarity. In addition, we can combine KPC clustering algorithm with thevoting strategy, the desired result of the validation experiments shows that the convergencerate is improved and enhance the clustering results accuracy.Finally, we know that the flood of information and the information Trek have troubledpeople for many years. To solve this, we design a personalized recommendation model basedthe improved clustering algorithm. What’s more, we present detail formal definition and listthe steps attached to the cluster center, simulation experiments result shows the feasibility of the process.