US 11,699,179 B2
Size and fitting recommendation systems and method for fashion products
Borar Sumit, Karnataka (IN); Ghani Mohammed Abdulla, Bangalore (IN); and Sengupta Srijit, Karnataka (IN)
Assigned to Myntra Designs Private Limited, Bangalore (IN)
Filed by Myntra Designs Private Limited, Bangalore (IN)
Filed on Feb. 4, 2019, as Appl. No. 16/266,789.
Claims priority of application No. 201841017210 (IN), filed on May 8, 2018.
Prior Publication US 2019/0347706 A1, Nov. 14, 2019
Int. Cl. G06Q 30/0601 (2023.01)
CPC G06Q 30/0631 (2013.01) 16 Claims
OG exemplary drawing
 
1. A size and fitting recommendation system for fashion products, the system comprising:
memory having computer-readable instructions stored therein; and
a processor configured to:
access purchase and content data of a plurality of first fashion products purchased by at least one user;
identify one or more second fashion products (i) having a second type with greater similarity to a first type of one or more of the first fashion products than with a third type of a third set of fashion products (ii) purchased less frequently than any of the first fashion products by a set of users, wherein the first and one or more second fashion products are previously purchased, and the greater similarity being based on a criterion being satisfied;
generate an observable feature vector for each of the first fashion products and another observable feature vector for each of the one or more second fashion products, wherein the observable feature vectors are generated based upon respective observable features data of the fashion products;
aggregate the observable feature vectors to compute an observable user vector;
generate a latent feature vector for each of the first fashion products and another latent feature vector for each of the one or more second fashion products, wherein the latent feature vectors are generated based upon respective latent features data of the fashion products;
aggregate the latent feature vectors to compute a latent user vector;
train an autoencoder model configured to (i) learn an initial compression, by densely representing the observable user vector and the latent user vector, (ii) learn a subsequent decompression, and (iii) generate at least one size and fitting recommendation for the one or more second fashion products, wherein the compression causes a reduction, during the training, of a number of dimensions of each of the observable user vector and the latent user vector; and
display, to the at least one user at a user interface, the at least one generated recommendation, comprising personalized size information across brands, fit type, brand type, product type, or combinations thereof of the one or more second fashion products.