Improving music artist recommendations through analysis of influences

To improve the quality of search results in huge digital music databases, we developed a simple algorithm based on artist influences and complex network theory that produces interesting and novel results. Traditionally, music recommendation engines use audio feature similarity to suggest new music based on a given artist. We propose a search that takes influences into account provides a richer result set than one based on audio features alone. We constructed an artist influence network using the Rovi dataset and studied it using complex network theory. Analysis revealed many complex network phenomena which we used to tune the search algorithm. Finally, we consider the difficulty of qualitatively rating our results and the need for a tool to exercise the algorithm.