US 12,248,974 B1
System and method for recommending resale alternatives for retail items
Phoebe Gates, New York, NY (US); Sophia Kianni, New York, NY (US); and Silas Alberti, Stanford, CA (US)
Assigned to Phia Holdings Inc., New York, NY (US)
Filed by Phia Holdings Inc., Menlo Park, CA (US)
Filed on Jan. 31, 2024, as Appl. No. 18/429,245.
Int. Cl. G06Q 30/00 (2023.01); G06F 40/20 (2020.01); G06Q 30/0601 (2023.01); G06V 20/50 (2022.01)
CPC G06Q 30/0631 (2013.01) [G06F 40/20 (2020.01); G06V 20/50 (2022.01); G06V 2201/10 (2022.01)] 17 Claims
OG exemplary drawing
 
9. A computer-implemented method for recommending resale consumer goods alternatives, comprising:
providing a recommendation algorithm to search and locate a resale consumer item from among one or more resale consumer item websites after receiving a request to search for a specific consumer item found on a consumer item website based on a subject image of the specific consumer item, wherein the resale consumer item is a secondhand good or a used item, and wherein the one or more resale consumer item websites provide the secondhand good or used item,
wherein the subject image of the specific consumer item is taken from the consumer item website and provided to the recommendation algorithm to conduct a search to find and locate a resale consumer item that is a nearest match to the specific consumer item;
using the recommendation algorithm, analyzing the subject image to determine one or more characteristics associated with the subject image further comprising:
extracting relevant product metadata and data from the subject image using metadata heuristics, large language model (LLM) based extraction, manual selector-database, and LLM-generated selector-database further comprising:
scraping product information from the consumer goods website by implementing two heuristics that extract the relevant metadata and data based on different common metadata standards in a header of the consumer goods website;
combining the extracted metadata from multiple sources into a single source; and
utilizing language models to extract structured JSON data from unstructured website content of the consumer goods website;
running the search in parallel using multiple product images, text vectors, and multiple filter sets to retrieve multiple result sets from the one or more resale consumer goods websites;
fusing and re-ranking the multiple result sets and obtaining multiple sets of scored results;
merging the multiple sets of scored results into a final result list of potential recommendations; and
transmitting the final result list of potential recommendations with the re-ranking to the user, wherein the final list of potential recommendations directs the user to a resale consumer item website that is separate from the consumer item website that was the source for the subject image.