This paper aims at analyzing user behavior from PC operation logs by the functionality clustering of networks. The functionality clustering method was proposed by Fushimi et al. This method employs PageRank algorithm and classifies nodes according to their functions and roles in the network. In this paper, to extract similar tasks as user behaviors from PC operation logs, we improve the conventional functionality clustering method by transforming PageRank convergence patterns into symbolic representations of time series data, and applying edit distance to these representations. We show that the improved method allows us to classify user behaviors according to their context at work.