公开(公告)号：US20210295185A1

公开(公告)日：2021-09-23

申请号：US17225056

申请日：2021-04-07

Applicant: Coupang Corp.

Inventor： Gil Ho LEE ,  Pankesh BAMOTRA

IPC: G06N5/04 ,  G06N20/00 ,  G06N5/02 ,  G06Q10/08 ,  G06K9/62 ,  G06F7/08 ,  G06F16/215

Abstract: Systems and methods are provided for categorizing products using AI. One method comprises retrieving initial training data including products associated with one or more categories; pre-processing the initial training data to generate synthesized training data; generating a hierarchical model using the synthesized training data, the hierarchical model containing at least two layers of nodes below a root node; receiving information associated with a first uncategorized product; and receiving a request to predict a set of N categories with the highest N total probability scores. The method may further comprise predicting, using the hierarchical model, N categories of the first uncategorized product, by calculating total probability scores, and determining the N categories with the highest N total probability scores; sorting the first uncategorized product into the N categories associated with the nodes from the first and second layers having the highest total probability scores; and displaying the sorted first uncategorized product and its associated N categories on a user device associated with a user.

公开(公告)号：US20220092666A1

公开(公告)日：2022-03-24

申请号：US17030282

申请日：2020-09-23

Applicant: Coupang, Corp.

Inventor： Nishant AGRAWAL ,  Pankesh BAMOTRA ,  Emily HUA ,  Ashutosh PENDSE ,  Sungho KIM ,  Chan SEO ,  Min Hyun PARK

IPC: G06Q30/06 ,  G06F16/2457

Abstract: One aspect of the present disclosure is directed to a computer-implemented system for streamlined product searching configured to receive a search query comprising at least one keyword; determine a product type; retrieve a first record of products associated with the product type; determine whether the search query meets a condition; and if the search query meets the condition: determine, a plurality of label values, and for each label value: generate a second record of products associated with the label value; determine whether each product in the second record meets a condition; deleting each product that meets the condition from the second record; ranking each remaining product in each second record; display the plurality of label values for selection; receive a selected label value; and display at least a portion of the remaining ranked products associated with the selected label value.