A hierarchical meta-model of user behaviors and a method for user behavior modeling based on the metadata in pages were proposed to overcome the difficulty of the existing methods in describing the semantics of behavior. In the method

the behavior model of web users is built with three levels

i.e.

URL accesses

activities and sessions. Furthermore

both the abstraction of metadata in pages and the transformation among URL accesses

activities and sessions are specified. The behavior models generated by the method can describe not only the information of user accessing sequence

but also the additional information including local themes of pages content and keywords. The information gives the foundation for acquiring the semantic characteristics of user behaviors. Tests performed in web-based distance learning system of Xi'an Jiaotong University show that the accuracy of activity partition by the proposed method exceeds 86%.