US 2026/0252571 A1Published application

Item retrieval using chain-of-thought query categorization

AssigneeeBay Inc.

InventorsJetlir DURAJ, Ishita Kamal KHAN, Kilian MERKELBACH, Mehran ELYASI

Life of this patent

Estimating the term…

Loading patent…

Abstract

Some aspects relate to technologies for a search engine that performs item retrieval on a listing platform using chain-of-thought (CoT) query categorization. The system receives a search query (e.g., a natural language query) and performs a tree search on a hierarchical taxonomy of categories using a language model that semantically scores category nodes. Less relevant branches are pruned, and traversal continues along high-scoring paths using different traversal approaches, such as a breadth-first or depth-first approach. This process identifies candidate leaf categories aligned with the search query. The language model then scores these candidate categories. Based on the relevance scores, one or more relevant leaf categories are selected. The system then queries an item listings data store constrained by the selected categories and returns the resulting search results.

Claims20

Dependent claims

One or more computer storage media storing computer-useable instructions that, when used by one or more computing devices, cause the one or more computing devices to perform operations, the operations comprising:

receiving a search query;
iterating through each level in a hierarchical taxonomy of categories until one or more leaf categories are reached, wherein each iteration comprises:
accessing category data for each category in a set of categories at a current level in the hierarchical taxonomy of categories,
causing a language model to generate a relevance score for each category in the set of categories using the search query and the category data for each category in the set of categories, wherein each relevance score indicates a relevance of a corresponding category from the set of categories to the search query,
selecting one or more categories from the set of categories using the relevance scores, wherein a next set of categories for a next level in the hierarchical taxonomy of categories comprises children categories of the selected one or more categories, and
when one or more leaf categories are reached, adding the one or more leaf categories to a set of candidate leaf categories;
causing the language model to generate a relevance score for each leaf category in the set of candidate leaf categories;
and
providing search results for the search query based on at least one leaf category selected from the set of candidate leaf categories based on the relevance score for each leaf candidate in the set of candidate leaf categories.

6 more elements

A computer-implemented method comprising:

receiving a search query;
iterating through subtrees in a hierarchical taxonomy of categories until one or more leaf categories are reached, wherein each iteration comprises:
selecting a subtree from the hierarchical taxonomy of categories,
causing a language model to traverse the subtree by generating a relevance score for each category in the subtree until the language model reaches an intermediate category that does not satisfy a threshold or the language model reaches one or more leaf categories in the subtree, wherein each relevance score indicates a relevance of a corresponding category from to the search query, and
when one or more leaf categories in the subtree are reached, adding the one or more leaf categories to a set of candidate leaf categories;
causing the language model to generate a relevance score for each leaf category in the set of candidate leaf categories;
and
providing search results for the search query based on at least one leaf category selected from the set of candidate leaf categories based on the relevance score for each leaf candidate in the set of candidate leaf categories.

5 more elements

Dependencies are inferred from the claim text.