A user behavior analysis is an important approach in many algorithms for the Web site information recommendation, among which, PageGather is a typical algorithm. However, the original PageGather algorithm is static, which needs too many data inputs and too much computing time. In this paper, incremental learning and distributed computation mechanisms are introduced into PageGather, so that two improved algorithms PG+ and PG++ are proposed. At the same time, corresponding experimental results are presented and analyzed.The improved algorithms are equivalent to the static PageGather algorithms.And better effect has got.