A Semantic-Based Approach for Artist Similarity

This paper describes and evaluates a method for computing artist similarity from a set of artist biographies. The proposed method aims at leveraging semantic information present in these biographies, and can be divided in three main steps, namely: (1) entity linking, i.e. detecting mentions to named entities in the text and linking them to an external knowledge base; (2) deriving a knowledge representation from these mentions in the form of a semantic graph or a mapping to a vector-space model; and (3) computing semantic similarity between documents. We test this approach on a corpus of 188 artist biographies and a slightly larger dataset of 2,336 artists, both gathered from Last.fm. The former is mapped to the MIREX Audio and Music Similarity evaluation dataset, so that its similarity judgments can be used as ground truth. For the latter dataset we use the similarity between artists as provided by the Last.fm API. Our evaluation results show that an approach that computes similarity over a graph of entities and semantic categories clearly outperforms a baseline that exploits word co-occurrences and latent factors.