Feature subspace isolation and disentanglement in merchant embeddings

Abstract

A computer-implemented method for providing merchant recommendations comprises receiving, by a processor, raw merchant embeddings generated from payment transaction records, wherein the raw merchant embeddings include a plurality of embedded features entangled in an embedding space. The processor uses transaction metadata associated with the payment transaction records to determine a subspace of an identified feature within the embedding space. A linear transformation process then removes the subspace of the identified feature from the embedding space to create modified merchant embeddings that are merged and aligned with other ones of the plurality of features within the embedding space. The processor automatically generates a list of merchant rankings based on the modified merchant embeddings, past preferences of a target user using raw user embeddings, and a target region, and provides the list of merchant rankings to the target user.