Building an Interactive Next-Generation Artist Recommender Based on Automatically Derived High-Level Concepts

We present a new way of accessing large sets of musical artists based on high-level concepts. The concepts are derived and assigned to individual artists by an automatic procedure: Using a list of music-related words and phrases, the well-known TF x IDF approach is applied to analyse the 100 top Web pages related to each artist, as delivered by a Web search engine. This data then is decomposed into a number of "archetypical" bases or "concepts" by non-negative matrix factorisation (NMF). Each artist is then described by the amount by which it is related to each of these concepts. In our browser application presented here, such a representation allows for independently adjusting the weight of each of these concepts, to recommend those artists that best match the desired query profile.