Query analysis using deep neural net classification

Patent metadata

Field	Value
Publication number	US-11074486-B2
Application number	US-201715822995-A
Country	US
Kind code	B2
Filing date	Nov 27, 2017
Priority date	Nov 27, 2017
Publication date	Jul 27, 2021
Grant date	Jul 27, 2021

Abstract

The present invention provides a method, computer program product, and system of generating predicted reactions of a user. In some embodiments, the method, computer program product, and system include receiving an intelligence data store, receiving a current data object with a current query and at least one knowledge graph, identifying one or more patterns in the at least one knowledge graph, comparing using a deep neural net, the previous queries and associated one or more patterns with the current query and identified one or more patterns of the current data object, classifying the plurality data objects from the intelligence data store based on a closeness of the current query and identified one or more patterns with each of the previous queries and associated one or more patterns in the intelligence data store, and identifying, by the classification engine, potential dispositions based on the classification of the plurality of data objects.

First claim

The invention claimed is:

1. A system comprising:
at least one processor;
an intelligence data store comprising a plurality of data objects comprising previous queries and associated one or more patterns with dispositions associated with the previous queries and associated one or more patterns;
a feature identifier configured, in response to receiving a current data object with a current query and at least one knowledge graph, to detect one or more patterns of entities and relations in the at least one knowledge graph,
wherein the current data object is related to a mobile phone;

a comparison engine configured to compare, using a deep neural net, the previous queries and associated one or more patterns with the current query and identified one or more patterns of the current data object,
wherein the comparing includes performing a graph traversal to identify a distance of the mobile phone from an incident;

a classification engine to score the plurality data objects from the intelligence data store with the current data object based on a closeness of the current query and identified one or more patterns with each of the previous queries and associated one or more patterns; and
an intelligence classification component configured to identify that the mobile phone was used at the time and place of an incident based on the scores of the plurality of data objects.

2. The system of claim 1 , wherein the patterns are one or more subgraphs and the classification engine is a neural net sub-graph classification engine.

3. The system of claim 2 , wherein the classification engine includes a feature layer which normalizes attributes of interest as features in a deep net.

4. The system of claim 3 , wherein the feature layer includes a vector representation and focusing algorithms.

5. The system of claim 3 , wherein the classification engine includes a 3+ layer multilayer perception (MLP).

6. The system of claim 3 , wherein the classification engine includes a convolutional neural net architecture to perform focus-shifting analysis of an overall graph.

7. The system of claim 3 , wherein the queries are from law enforcement.

8. The system of claim 3 , wherein the queries are from an intelligence agency.

9. The system of claim 3 , wherein the feature layer includes a subgraph normalization to vector representation.

10. The system of claim 1 , wherein the one or more patterns are classified with labels.

11. The system of claim 1 further comprising, a reporting engine, configured to report that the mobile phone was used at the time and place of an incident via a display logically connected to the system.

12. The method of claim 1 , wherein the comparing further includes performing graph traversal measures hops in the vicinity of the incident.

13. A method for answering a query comprising:
receiving an intelligence data store comprising a plurality of data objects comprising previous queries and associated one or more patterns with dispositions associated with the previous queries and associated one or more patterns;
receiving a current data object with a current query and at least one knowledge graph;
identifying, by a feature identifier, one or more patterns of entities and relations in the at least one knowledge graph,
wherein the current data object is related to a mobile phone;
comparing, by a comparison engine, using a deep neural net, the previous queries and associated one or more patterns with the current query and identified one or more patterns of the current data object,
wherein the comparing includes performing a graph traversal to identify a distance of the mobile phone from an incident;
classifying, by a classification engine, the plurality data objects from the intelligence data store based on a closeness of the current query and identified one or more patterns with each of the previous queries and associated one or more patterns; and
identifying, by the classification engine, that the mobile phone was used at the time and place of an incident based on the classification of the plurality of data objects.

14. The method of claim 13 , wherein the patterns are one or more subgraphs and the classification engine is a neural net sub-graph classification engine.

15. The method of claim 13 , wherein the classification engine includes a feature layer which normalizes attributes of interest as features in a deep net.

16. The method of claim 15 , wherein the feature layer includes a vector representation and focusing algorithms.

17. The method of claim 15 , wherein the classification engine includes a 3+ layer multilayer perception (MLP).

18. The method of claim 15 , wherein the classification engine includes a convolutional neural net architecture to perform focus-shifting analysis of an overall graph.

19. The method of claim 15 , wherein the queries are from law enforcement.

20. A computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to perform a method comprising:
receiving an intelligence data store comprising a plurality of data objects comprising previous queries and associated one or more patterns with dispositions associated with the previous queries and associated one or more patterns;
receiving a current data object with a current query and at least one knowledge graph;
identifying, by a feature identifier, one or more patterns of entities and relations in the at least one knowledge graph,
wherein the current data object is related to a mobile phone;
comparing, by a comparison engine, using a deep neural net, the previous queries and associated one or more patterns with the current query and identified one or more patterns of the current data object,
wherein the comparing includes performing a graph traversal to identify a distance of the mobile phone from an incident;
classifying, by a classification engine, the plurality data objects from the intelligence data store based on a closeness of the current query and identified one or more patterns with each of the previous queries and associated one or more patterns; and
identifying, by the classification engine, that the mobile phone was used at the time and place of an incident based on the classification of the plurality of data objects.

Assignees

• IBM

Inventors

• Linton Jeb R
• Wright David K

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06F16/2457 Primary

  with adaptation to user needs · CPC title

• G06V30/1988 Primary

  Graph matching (graphical image representation G06V30/18181) · CPC title