US 2025/0061136 A1Published application

Information Retrieval Using an Augmented Query Produced by Graph Convolution

AssigneeMicrosoft Technology Licensing, LLC

InventorsDeepak SAINI, Jian JIAO, Aditya Lakshman KALARI, Chaoyin LI

Life of this patent

Estimating the term…

Loading patent…

Abstract

An information retrieval technique uses one or more machine-trained models to generate one or more metadata embeddings. The technique then combines a query embedding with the metadata embedding(s). In some cases, the technique performs this operation using a graph convolution operation. This yields an augmented embedding. The technique then uses the augmented embedding to retrieve at least one item. The augmented embedding lies in the same vector space as target-item embeddings associated with candidate target items. Otherwise, the vector spaces associated with the query embedding and metadata embedding(s) can be different. In some implementations, the technique use dense retrieval, which enables the technique to deliver output results in real time.

Claims20

Dependent claims

A method for identifying one or more target items that match an input query, comprising:

receiving the input query;
encoding the input query into a query embedding using a first machine-trained model that is trained using a first set of training examples, the query embedding being a vector in a first vector space;
producing a metadata embedding based on the input query using a second machine-trained model that is produced using a second set of training examples, the metadata embedding being associated with a particular instance of metadata and being a vector in a second vector space;
combining the query embedding with the metadata embedding using a third machine-trained model that is produced using a third set of training examples, to produce an augmented embedding, the augmented embedding being a vector in a third vector space;
identifying a target item based on the augmented embedding;
and
generating output information for presentation by an output device, the output information including the target item or a reference to the target item,
the first vector space, second vector space, and third vector space being three different and independent vector spaces, and the first set of training examples, the second set of training examples, and the third set of training examples being, at least in part, different sets of training examples.

5 more elements

A computing system for identifying one or more target items that match an input query, comprising:

an instruction data store for storing computer-readable instructions;
a processing system for executing the computer-readable instructions in the data store, to perform operations including:
receiving the input query;
encoding the input query into a query embedding using a first machine-trained model;
producing a metadata embedding based on the input query using a second machine-trained model, the metadata embedding being associated with a particular instance of metadata;
in a graph convolution operation performed using a third machine-trained model, merging the query embedding with the metadata embedding, to produce a merged embedding, and transforming the merged embedding to an augmented embedding;
searching a target-item index to find a target-item embedding that is closest to the augmented embedding, the target item embedding being associated with a target item;
and
generating output information for presentation by an output device, the output information including the target item or a reference to the target item.

6 more elements

Dependencies are inferred from the claim text.