Abstract-This paper aims to discriminate user behavior from PC operation logs by PageRank convergence patterns of networks. We will show that we can discriminate user behavior by making clear the difference in behavior between working and resting. We will construct a window transition network from active window transition logs. In this network, each node represents an application corresponding to the active window. We will use PageRank convergence patterns of networks. The convergence patterns are assumed to imply the roles of nodes in the network. Then, we will transform these patterns into symbolic representations to compute similarities in user behavior and apply the kernel method for classification with SVMs. We will conduct experimental result. This evaluation is highly accurate except amongst people who seldom use computers. These results show that this method allows us to discriminate user behavior according to the context at work with SVMs.