To solve the contradiction between massive network information and limited learning needs, personalized recommendation service becomes the hotspot in research area. User characteristics analysis is the key point in personalized recommendation service. Based on the massive web access log in the web server, this research gradually puts forward the steps of user characteristics analysis which including data pretreatment, user feature extraction and user clustering. This paper focuses on the rules of user recognition, definition of user feature, user feature extraction algorithm and user group clustering. Finally, take access log files of a web server as sample, simulation experiments have been made to prove the thought put forward by this context. Improvement has been token to the push service repository on the basis of the experimental results, which achieved good practical results.