Vector-based search result generation

Patent metadata

Field	Value
Publication number	US-12079185-B2
Application number	US-202017084201-A
Country	US
Kind code	B2
Filing date	Oct 29, 2020
Priority date	Oct 29, 2020
Publication date	Sep 3, 2024
Grant date	Sep 3, 2024

Abstract

A system and method to generate search results in response to a search query based on comparisons of embedding vectors. The system and method receive, from an end user system, a search query including a set of keywords associated with the entity. Using a neural network, an embedding vector is identified based on the set of keywords of the search query. The system and method compares the embedding vector associated with the search query to a set of embedding vectors associated with a set of structured data elements relating to the entity. Based on the comparison, a set of matching structured data elements is identified. The system and method generate a search result in response to the search query, wherein the search result includes at least a portion of the set of matching structured data elements. The search result is displayed via an interface of the end user system.

First claim

What is claimed is:

1. A method comprising:
selecting, during a training phase associated with a neural network, from one or more search logs, a set of pairs, wherein each pair comprises:
a query; and
a previously answered frequently asked question associated with the query;

generating, by the neural network executed by a processing device, a set of embedding vectors for each pair in the set of pairs, resulting in a plurality of embedding vectors for previous queries, wherein each set of embedding vectors comprises an embedding of the query and an embedding of the previously answered frequently asked question associated with the query;
generating an entity-specific index, wherein the entity-specific index comprises a set of indices for the plurality of embedding vectors for previous queries;
receiving, from an end user system, a search query associated with an entity associated with the entity-specific index;
generating, by the neural network, at least one query embedding based on the search query;
comparing, using the entity-specific index, the at least one query embedding to the plurality of embedding vectors for previous queries using a distance measurement, resulting in a ordered list of frequently asked questions corresponding to the search query based on the distance measurement between the at least one query embedding and the plurality of embedding vectors for previous queries; and
generating a search result in response to the search query, wherein the search result comprises the frequently asked question within the ordered list having a smallest distance.

2. The method of claim 1 , further comprising generating, based on the comparing, scores representing a level of matching between the the at least one query embedding and the plurality of embedding vectors for previous queries.

3. The method of claim 1 , further comprising:
causing a display of the search result via an interface of the end user system.

4. The method of claim 1 , further comprising:
causing a display of the search result comprises a ranked listing of the frequently asked questions based on the ordered list.

5. A system comprising:
a memory to store instructions; and
a processing device, operatively coupled to the memory, to execute the instructions to perform operations comprising:
selecting, during a training phase associated with a neural network, from one or more search logs, a set of pairs, wherein each pair comprises:
a query; and
a previously answered frequently asked question associated with the query;

generating, by the neural network executed by a processing device, a set of embedding vectors for each pair in the set of pairs, resulting in a plurality of embedding vectors for previous queries, wherein each set of embedding vectors comprises an embedding of the query and an embedding of the previously answered frequently asked question associated with the query;
generating an entity-specific index, wherein the entity-specific index comprises a set of indices for the plurality of embedding vectors for previous queries;
receiving, from an end user system, a search query associated with an entity associated with the entity-specific index;
generating, by the neural network, at least one query embedding based on the search query;
comparing, using the entity-specific index, the at least one query embedding to the plurality of embedding vectors for previous queries using a distance measurement, resulting in a ordered list of frequently asked questions corresponding to the search query based on the distance measurement between the at least one query embedding and the plurality of embedding vectors for previous queries; and
generating a search result in response to the search query, wherein the search result comprises the frequently asked question within the ordered list having a smallest distance.

6. The system of claim 5 , the operations further comprising:
storing, in a data store, the entity-specific index.

7. The system of claim 5 , the operations further comprising:
generating scores representing a level of matching between the the at least one query embedding and the plurality of embedding vectors for previous queries.

8. The system of claim 7 , wherein the display of the search result comprises a ranked listing of the frequently asked questions based on the ordered list.

9. The system of claim 7 , the operations further comprising:
identifying a first subset of the set of matching structured data elements having a corresponding score that is less than a threshold level; and
removing the first subset from the search result.

10. A non-transitory computer readable storage medium comprising instructions that, when executed by a processing device, cause the processing device to perform operations comprising:
selecting, during a training phase associated with a neural network, from one or more search logs, a set of pairs, wherein each pair comprises:
a query; and
a previously answered frequently asked question associated with the query;

generating, by the neural network executed by a processing device, a set of embedding vectors for each pair in the set of pairs, resulting in a plurality of embedding vectors for previous queries, wherein each set of embedding vectors comprises an embedding of the query and an embedding of the previously answered frequently asked question associated with the query;
generating an entity-specific index, wherein the entity-specific index comprises a set of indices for the plurality of embedding vectors for previous queries;
receiving, from an end user system, a search query associated with an entity associated with the entity-specific index;
generating, by the neural network, at least one query embedding based on the search query;
comparing, using the entity-specific index, the at least one query embedding to the plurality of embedding vectors for previous queries using a distance measurement, resulting in a ordered list of frequently asked questions corresponding to the search query based on the distance measurement between the at least one query embedding and the plurality of embedding vectors for previous queries; and
generating a search result in response to the search query, wherein the search result comprises the frequently asked question within the ordered list having a smallest distance.

11. The non-transitory computer readable storage medium of claim 10 , the operations further comprising generating, based on the comparing, scores representing a level of matching between the the at least one query embedding and the plurality of embedding vectors for previous queries.

12. The non-transitory computer readable storage medium of claim 11 , wherein the search result comprises a ranked listing of the frequently asked questions based on the ordered list.

13. The non-transitory computer readable storage medium of claim 12 , the operations further comprising:
identifying a first portion of the set of matching structured data elements having a corresponding score that is less than a threshold level; and
removing the first portion from the search result.

Assignees

• Yext Inc

Inventors

• Misiewicz Michael
• Dunn Michael
• Davish Maxwell

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06F16/287

  Visualization; Browsing · CPC title

• G06F16/288

  Entity relationship models · CPC title

• G06N5/02

  Knowledge representation; Symbolic representation · CPC title