In this paper, we propose a graph-based framework to organize low-level and high-level features of music objects in a unified way. The featured graph, called the power graph, is associated with operators to support a variety of music information retrieval applications, such as auto-tagging, link analysis, similarity measurement, and clustering. Among these operators, we have identified the node ranking by computing prestige value as one of the essential fundamental link analysis operators. For this particular operator, we propose two methods of computing prestige; they are the power method and the algebraic method. Although the algebraic method is originated from the symmetric graph, the algebraic method can be applied as an approximate but efficient alternative to the power method. To demonstrate the feasibility of our framework, we have carried out an auto-tagging experiment and a music object clustering experiment. According to the auto-tagging experimental results, we have observed that the algebraic method has achieved almost the same results as the power method with only a one-fifth elapsed time. In the experiments we have conducted, we have achieved accuracy levels up to 75 %. © 2014 Springer Science+Business Media New York.