Finding Concepts of Music Objects with Unexpected Multi-labels Based on Shared Subspace Method

We discuss in this paper a recommendation-oriented information retrieval for music data. Particularly, we present a general framework of retrieving unexpected objects for a given query, where each data object is represented as a feature vector and assigned a multi-label as well. Given an object-feature matrix X1 and an object-label matrix X2, we simultaneously factorize X1 and X2 as X1 is approximately equal to B V and X2 to S W by means of Nonnegative Shared Subspace Method, where the basis S is a part (subspace) of the basis B. Such a shared subspace associates the label-information with the feature-information of the original matrices. Based on the shared subspace, thus, we can predict a multi-label for a query feature-vector with unknown labels. Our unexpected object for the query is defined as an object which is similar to the query in the feature space, but is apart from the query in the label space. In order to obtain those unexpected objects from several viewpoints of similarity, we formalize our retrieval task as a problem of finding formal concepts satisfying a constraint w.r.t. the unexpectedness. Our experimental result for a dataset of music pieces created from Million Song Dataset Benchmarks shows we can actually detect an interesting music cluster as the extent of a formal concept including some unexpected music pieces for a given music query.