The user-assigned tag is a growingly important research topic in MIR. Noticing that some tags are more specific versions of others, this paper studies the problem of organizing tags into a hierarchical structure by taking into ac- count the fact that the corresponding artists are organized into a hierarchy based on genre and style. A novel clustering algorithm, Hierarchical Co-clustering Algorithm (HCC), is proposed as a solution. Unlike traditional hierarchical clustering algorithms that deal with homogeneous data only, the proposed algorithm simultaneously organizes two distinct data types into hierarchies. HCC is additionally able to receive constraints that state certain objects "must-be-together" or "should-be-together" and build clusters so as to satisfying the constraints. HCC may lead to better and deeper understandings of relationship between artists and tags assigned to them. An experiment finds that by trying to hierarchically cluster the two types of data better clusters are obtained for both. It is also shown that HCC is able to incorporate instance-level constraints on artists and/or tags to improve the clustering process.