Monomer: Non-Metric Mixtures-of-Embeddings for Learning Visual Compatibility Across Categories.
Ruining He, Charles Packer, Julian McAuley · arXiv (Cornell University) · 2016
Identifying relationships between items is a key task of an online recommender system, in order to help users discover items that are functionally complementary or visually compatible. In domains like clothing recommendation, this task is particularly challenging since a successful system should be capable of handling a large corpus of items, a huge amount of relationships among them, as well as the high-dimensional and semantically complicated features involved. Furthermore, the human notion of that we need to capture goes beyond mere similarity: For two items to be compatible---whether jeans and a t-shirt, or a laptop and a charger---they should be similar in some ways, but systematically different in others. In this paper we develop a method, Monomer, to uncover complicated and heterogeneous of relationships between items. Recently, scalable methods have been developed that address this task by learning embeddings of the visual and textual characteristics of the products involved, but which ultimately depend on a nearest-neighbor assumption between the learned embeddings. Here we show that richer notions of compatibility can be learned, principally by relaxing the metricity assumption inherent in previous work, so as to uncover ways in which related items should be systematically similar, and systematically different. Quantitatively, we show that our system achieves state-of-the-art performance on large-scale compatibility prediction tasks, especially in cases where there is substantial heterogeneity between related items.