US 11,915,289 B2
Query reformulations for an item graph
Jonathan Lennart Bender, San Francisco, CA (US); Tyler Russell Tate, Vallejo, CA (US); Tejaswi Tenneti, Fremont, CA (US); and Aditya Subramanian, San Francisco, CA (US)
Assigned to Maplebear Inc., San Francisco, CA (US)
Filed by Maplebear, Inc., San Francisco, CA (US)
Filed on Mar. 1, 2021, as Appl. No. 17/188,214.
Prior Publication US 2022/0277373 A1, Sep. 1, 2022
Int. Cl. G06Q 30/00 (2023.01); G06Q 30/0601 (2023.01); G06F 16/901 (2019.01); G06F 17/18 (2006.01)
CPC G06Q 30/0625 (2013.01) [G06F 16/9024 (2019.01); G06F 17/18 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A method, performed at a computer system comprising a processor and a computer-readable medium, comprising:
receiving a search query including one or more search terms from a customer via a user interface of the computer system;
retrieving an item graph comprising item nodes and attribute nodes, wherein the item graph is a database of records arranged as a graph database, and each attribute node is connected to a set of item nodes that share an attribute represented by the attribute node;
segmenting the search query into tokens, each token comprising one or more of the search terms;
identifying candidate nodes from the item graph by comparing the tokens to the item graph;
in response to determining that none of the identified candidate nodes meet presentation criteria, reformulating the search query by applying a query reformulation model to the one or more search terms, wherein the query reformulation model calculates a first probability that the search query leads to a first conversion and a second probability that a reformulated search query leads to a second conversion, and the second probability is higher than the first probability, and wherein the query reformulation model is trained with a dataset including the search terms entered by the customer, subsequent search terms entered by customers, and labels indicating whether the subsequent search terms resulted in a conversion;
selecting search results from the item graph comprising one or more item nodes based on the reformulated search query; and
transmitting the search results to a customer device.