US 12,248,872 B1
Machine learning framework for personalized clothing compatibility
Anurag Beniwal, Redmond, WA (US); Meet Taraviya, Los Angeles, CA (US); Yen-Liang Lin, Aliso Viejo, CA (US); and Larry Davis, Brooklyn, NY (US)
Assigned to Amazon Technologies, Inc., Seattle, WA (US)
Filed by Amazon Technologies, Inc., Seattle, WA (US)
Filed on Sep. 27, 2021, as Appl. No. 17/486,666.
Claims priority of provisional application 63/231,574, filed on Aug. 10, 2021.
Int. Cl. G06N 3/08 (2023.01); G06F 18/21 (2023.01); G06Q 10/083 (2023.01)
CPC G06N 3/08 (2013.01) [G06F 18/21 (2023.01); G06Q 10/083 (2013.01)] 18 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
for each individual clothing item of a first plurality of clothing items:
generating a first user interface that includes an image of the individual clothing item and prompts a user for input regarding whether the user likes the individual clothing item; and
receiving, via user interaction with the first user interface, a first indication of whether the user likes the individual clothing item;
based at least in part on user input provided via the first user interface for the individual clothing items in the first plurality of clothing items, generating a user-specific embedding associated with the user for use within a machine learning model, wherein the machine learning model is trained to receive a proposed pair of clothing items and to output a predicted pairwise rating for the proposed pair of clothing items, wherein training data used in training the machine learning model includes feedback provided from a plurality of users with respect to clothing item pairs;
providing inputs to the machine learning model to generate a pairwise rating for a first clothing item and a second clothing item with respect to the user, wherein the inputs to the machine learning model comprise at least an image of the first clothing item and an image of the second clothing item;
generating, using the machine learning model, a plurality of subspace embeddings for each of the first clothing item and the second clothing item;
generating, using the machine learning model, a user-specific subspace embedding for each of the first clothing item and the second clothing item by weighting individual subspaces of the plurality of subspace embeddings based at least in part on the user-specific embedding;
generating, using the machine learning model, the pairwise rating representing user-specific compatibility of the first clothing item and the second clothing item being worn together, wherein the pairwise rating is specific to the user based at least in part on use of the user-specific subspace embedding for the first clothing item and the second clothing item; and
based at least in part on the pairwise rating, generating an outfit recommendation for presentation to the user, wherein the outfit recommendation suggests that the user wear the first clothing item and the second clothing item together.