Outfit recommender system
An outfit recommender system includes an outfit recommender and a storage system. The storage system stores embedding information for available inventory and for preconfigured outfits. Available inventory changes dynamically as items are added or sold by a provider. Preconfigured items are grouped together to form outfits and are generated from any source regardless of whether such items are available or exist in the market. After a customer selects an item to view, a preconfigured item most visually similar to the selected item is obtained. The preconfigured item has associated preconfigured complementary items that together form the preconfigured outfit. Next, complementary items in available inventory most visually similar to the preconfigured complementary items are obtained and presented to the customer. The novel outfit recommender provides dynamic outfit recommendations that change as available inventory changes based on static, pre-curated outfits. Complementary items are optionally filtered based on filter and/or profile information.
1 . A method comprising:
generating a plurality of pre-curated outfits, wherein each of the pre-curated outfits includes complementary items;
generating complementary item embeddings for each of the pre-curated outfits using an embedding model;
identifying taxons of each complementary item embedding;
labeling complementary item embeddings with taxon identifiers based on identified taxons;
generating embeddings for each available item physically in available inventory, wherein a provider entity sells items in available inventory via an online commerce platform;
continuously updating available item embeddings for each available item in available inventory as items are added to or removed from available inventory, wherein each available item embedding represents visual characteristics of an associated item;
continuously updating labeling of available item embeddings with taxon identifiers as items are added to or removed from available inventory;
obtaining a selected item embedding associated with a selected item in available inventory, wherein the selected item is selected via the online commerce platform, and wherein the selected item has a selected item taxon;
identifying a preconfigured item most visually similar to the selected item by comparing via cosine similarity the selected item embedding to preconfigured item embeddings with matching taxon identifiers, wherein the preconfigured item is associated with preconfigured complementary items, and wherein the preconfigured item and the associated preconfigured complementary items together represent a pre-curated outfit; and
obtaining complementary items in available inventory most visually similar to the preconfigured complementary items, wherein the complementary items are obtained by comparing via cosine similarity each of the preconfigured complementary item embeddings to available inventory item embeddings with matching taxon identifiers.
2 . The method of claim 1 , wherein the available inventory is stored in a storage system, wherein the preconfigured item and preconfigured complementary items are part of one of many preconfigured outfits stored in the storage system, and wherein the preconfigured outfits and the available inventory are represented as embeddings stored in the storage system.
3 . The method of claim 1 , further comprising:
presenting the complementary items on a product detail page (PDP) along with the selected item, wherein the PDP is dynamically updated based on changes in inventory availability of the complementary items.
4 . The method of claim 1 , wherein the preconfigured item and preconfigured complementary items are not required to be available in inventory to obtain the complementary items in available inventory.
5 . The method of claim 1 , wherein the embedding is a low-dimensional vector, wherein the embedding represents one or more of image information, item description information, taxon information, price information, provider or source information, sale information, market characteristic information, or other filter information.
6 . The method of claim 1 , further comprising:
filtering complementary items in available inventory based on filter information to obtain filtered complementary items; and
presenting the filtered complementary items.
7 . The method of claim 6 , wherein the filter is selected from the group consisting of: a price filter, a designer type filter, a time filter, a demand filter, and a preconfigurable filter.
8 . A system comprising:
a storage system, wherein the storage system stores embeddings for available inventory and embeddings for preconfigured outfits; and
an outfit recommender, wherein the outfit recommender is configured to: (i) obtain an embedding associated with a selected item in available inventory, (ii) obtain a preconfigured item most visually similar to the selected item via cosine similarity comparison only with preconfigured item embeddings having matching pre-labeled taxons, wherein the preconfigured item has associated preconfigured complementary items, and (iii) obtain complementary items in available inventory most visually similar to the preconfigured complementary items via cosine similarity comparison of each preconfigured complementary item embedding only with available inventory embeddings having matching pre-labeled taxons.
9 . The system of claim 8 , further comprising:
a product detail page (PDP), wherein an identifier is received onto the outfit recommender in response to a user selecting to view the PDP for the selected item, and wherein the complementary items are presented on the PDP along with the selected item.
10 . The system of claim 8 , wherein the outfit recommender handles communication in accordance with a representational state transfer application programming interface (REST API).
11 . The system of claim 8 , wherein the preconfigured item and preconfigured complementary items are not required to be available in inventory to obtain the complementary items in available inventory.
12 . The system of claim 8 , wherein the embedding is a low-dimensional vector, wherein the embedding represents one or more of image information, taxon information, price information, provider information, sale information, or filter information.
13 . A method comprising:
tracking user activity and market trends on an electronic commerce platform operated by an online consignment entity that operates a two-sided marketplace of buyers and sellers;
generating a plurality of pre-curated outfits, wherein each of the pre-curated outfits includes complementary items;
generating complementary item embeddings for each of the pre-curated outfits using an embedding model;
labeling complementary item embeddings with taxon identifiers based on identified taxons;
generating embeddings for each available item physically in available inventory;
continuously updating available item embeddings for each available item in available inventory as items are added to or removed from available inventory, wherein each available item embedding represents visual characteristics of an associated item;
continuously updating labeling of available item embeddings with taxon identifiers as items are added to or removed from available inventory;
obtaining a selected item embedding associated with a selected item in available inventory, wherein the selected item is selected via the online commerce platform, and wherein the selected item has a selected item taxon;
identifying a preconfigured item most visually similar to the selected item by comparing via cosine similarity the selected item embedding to preconfigured item embeddings with matching taxon identifiers, wherein the preconfigured item is associated with preconfigured complementary items, and wherein the preconfigured item and the associated preconfigured complementary items together represent a pre-curated outfit;
obtaining complementary items in available inventory most visually similar to the preconfigured complementary items, wherein the complementary items are obtained by comparing via cosine similarity each of the preconfigured complementary item embeddings to available inventory item embeddings with matching taxon identifiers; and
filtering the complementary items based on tracked user activity and market trends.
14 . The method of claim 13 , further comprising:
presenting filtered complementary items to a user of the electronic commerce platform.