Web Usage Mining is one of the important methods for web recommendations, but most of its studies are limited in using web server log, and its applications are limited in serving a particular web site. In this paper, based on mining the enterprise proxy log, we propose a novel WWW-oriented web recommendation system. Unlike other data sources, the enterprise proxy log is access history of visiting the whole World Wide Web. We firstly compare the differences between the web server log and the enterprise proxy log, and then an incremental data cleaning algorithm is proposed based on these differences. In the step of data mining, we present a clustering algorithm using hierarchical URL similarity. Experimental results show that this system can apply the technology of Web Usage Mining successfully in this new area.