Search platform for unstructured interaction summaries

Patent metadata

Field	Value
Publication number	US-12124487-B2
Application number	US-202318129997-A
Country	US
Kind code	B2
Filing date	Apr 3, 2023
Priority date	May 13, 2021
Publication date	Oct 22, 2024
Grant date	Oct 22, 2024

Abstract

Systems, methods, and computer program products for search platforms for unstructured interaction summaries. An application executing on a processor may receive a query comprising a term. The application may generate, based on an embedding vector and the term, an expanded query comprising a plurality of additional terms. The application may generate, based on a term frequency inverse document frequency model, a vector for the expanded query and generate an entity vector for the query. The application may generate a combined vector for the query based on the entity vector and the vector for the expanded query. The application may compute, based on the combined vector for the query and a feature matrix of a corpus, a respective cosine similarity score for a plurality of results in the corpus. The application may return one or more of the plurality of results as responsive to the query based on the similarity scores.

First claim

What is claimed is:

1. A computer-implemented method, comprising:
accessing, by an application executing on a processor, a feature matrix for a corpus comprising a plurality of text summaries, wherein the feature matrix represents each term in the plurality of text summaries as a respective feature of a plurality of features, and wherein the feature matrix indicates whether each of a plurality of entities is present in the respective text summary;
receiving, by the application, a query comprising a term;
generating, by the application based on an embedding vector and the term, an expanded query comprising a plurality of additional terms and the term;
generating, by the application based on a term frequency-inverse document frequency (TF-IDF) model, a vector for the expanded query;
generating, by the application, an entity vector for the query;
generating, by the application, a combined vector for the query based on the entity vector and the vector for the expanded query;
computing a respective score for each respective result of a plurality of results from the corpus; and
returning, by the application based on the scores, at least one of the plurality of results responsive to the query.

2. The computer-implemented method of claim 1 , wherein the scores comprise cosine similarity scores.

3. The computer-implemented method of claim 2 , wherein the cosine similarity scores are computed based on a product of the combined vector for the query and at least a portion of the feature matrix of the corpus.

4. The computer-implemented method of claim 1 , wherein a plurality of values of the embedding vector are trained based on the corpus, wherein the plurality of text summaries comprises unstructured text.

5. The computer-implemented method of claim 1 , wherein the combined vector for the query comprises a plurality of features, the method further comprising:
receiving, by the application, input labeling a first feature of the plurality of features as relevant to the query;
receiving, by the application, input labeling a second feature of the plurality of features as not relevant to the query;
removing, by the application, the second feature from the combined vector for the query; and
updating, by the application, the combined vector based on the remaining plurality of features and a respective weight for each remaining feature.

6. The computer-implemented method of claim 1 , further comprising prior to generating the expanded query:
preprocessing, by the application, the query to convert the query from a first format to a second format.

7. A non-transitory computer-readable storage medium, the computer-readable storage medium including instructions that when executed by a processor, cause the processor to:
access a feature matrix for a corpus comprising a plurality of text summaries, wherein the feature matrix represents each term in the plurality of text summaries as a respective feature of a plurality of features, and wherein the feature matrix indicates whether each of a plurality of entities is present in the respective text summary;
receive a query comprising a term;
generate, based on an embedding vector and the term, an expanded query comprising a plurality of additional terms and the term;
generate, based on a term frequency-inverse document frequency (TF-IDF) model, a vector for the expanded query;
generate an entity vector for the query;
generate a combined vector for the query based on the entity vector and the vector for the expanded query;
compute a respective score for each respective result of a plurality of results from the corpus; and
return, based on the scores, at least one of the plurality of results as responsive to the query.

8. The computer-readable storage medium of claim 7 , wherein the scores comprise cosine similarity scores.

9. The computer-readable storage medium of claim 8 , wherein the cosine similarity scores are computed based on a product of the combined vector for the query and at least a portion of the feature matrix of the corpus.

10. The computer-readable storage medium of claim 7 , wherein a plurality of values of the embedding vector are trained based on the corpus, wherein the plurality of text summaries comprises unstructured text.

11. The computer-readable storage medium of claim 7 , wherein the combined vector for the query comprises a plurality of features, wherein the instructions further cause the processor to:
receive input labeling a first feature of the plurality of features as relevant to the query;
receive input labeling a second feature of the plurality of features as not relevant to the query;
remove the second feature from the combined vector for the query; and
update the combined vector based on the remaining plurality of features and a respective weight for each remaining feature.

12. The computer-readable storage medium of claim 7 , wherein the instructions further cause the processor to, prior to generating the expanded query:
preprocess the query to convert the query from a first format to a second format.

13. A computing apparatus comprising:
a processor; and
a memory storing instructions that, when executed by the processor, cause the processor to:
access a feature matrix for a corpus comprising a plurality of text summaries, wherein the feature matrix represents each term in the plurality of text summaries as a respective feature of a plurality of features, and wherein the feature matrix indicates whether each of a plurality of entities is present in the respective text summary;
receive a query comprising a term;
generate, based on an embedding vector and the term, an expanded query comprising a plurality of additional terms and the term;
generate, based on a term frequency-inverse document frequency (TF-IDF) model, a vector for the expanded query;
generate an entity vector for the query;
generate a combined vector for the query based on the entity vector and the vector for the expanded query;
compute a respective score for each respective result of a plurality of results from the corpus; and
return, based on the scores, at least one of the plurality of results as responsive to the query.

14. The computing apparatus of claim 13 , wherein the scores comprise cosine similarity scores.

15. The computing apparatus of claim 14 , wherein the cosine similarity scores are computed based on a product of the combined vector for the query and at least a portion of the feature matrix of the corpus.

16. The computing apparatus of claim 13 , wherein a plurality of values of the embedding vector are trained based on the corpus, wherein the plurality of text summaries comprises unstructured text.

17. The computing apparatus of claim 13 , wherein the instructions further cause the processor to, prior to generating the expanded query: preprocess the query to convert the query from a first format to a second format.

18. The method of claim 1 , wherein the scores are computed based on the combined vector for the query and at least a portion of the feature matrix of the corpus.

19. The computer-readable storage medium of claim 7 , wherein the scores are computed based on the combined vector for the query and at least a portion of the feature matrix of the corpus.

20. The computing apparatus of claim 13 , wherein the scores are computed based on the combined vector for the query and at least a portion of the feature matrix of the corpus.

Assignees

• Capital One Services Llc

Inventors

• Ramamohan Deepak

Classifications

• G06F16/3347

  using vector based model · CPC title

• G06F16/338

  Presentation of query results · CPC title

• G06N20/00

  Machine learning · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title

• G06F16/3338 Primary

  Query expansion · CPC title