We propose a new classification system based on an analysis of folksonomy data. To find valuable resources from current social book-mark services, users need to specify search terms or tags, or to discover people with similar interests. Our system uses semantic relationships extracted from the co-occurrences of folksonomy data using PLSI and allocates folksonomy tags in a directed acyclic graph. Compared to the hierarchical allocation method of a tree, our method guarantees the number of children nodes and increases the number of available paths to an objective node, enabling users to navigate the resources using tags.