Sensor based context augmentation of search queries

• US 10,275,519 B2
• Filed: 08/22/2016
• Issued: 04/30/2019
• Est. Priority Date: 08/22/2016
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Cohn, John M., Kozhaya, Joseph N., Pureswaran, Veena S.
• Primary Examiner(s): Alam, Hosain T
• Assistant Examiner(s): May, Robert F
• Application Number:
  US15/243,729
• Publication Number:
  US 20180052915A1
• Time in Patent Office: 981 Days
• Field of Search: 707723
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

A computing device comprising a processor configured to receive a natural language search query and determine a first response to the natural language search query, the first response having a first confidence level. The processor is further configured to determine a sensor class effecting an information gain on the natural language search query according to a table of training data. The processor is further configured to augment the natural language search query to form an augmented search query according to data received from at least one sensor that is a member of the sensor class and determine a second response to the augmented search query, the second response having a second confidence level. The processor is further configured to store information to the table of training data that associates the at least one sensor with a value of a change between the first confidence level and the second confidence level.

Claims

1. A computer program product for sensor augmented searching, the 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:

receive a natural language search query;

determine a first response to the natural language search query, the first response having a first confidence level;

determine a sensor class that effects an information gain on the natural language search query according to a table of training data;

augment the natural language search query to form an augmented search query according to data received from at least one sensor that is a member of the sensor class by;

determining, based on the table of training data relating relevance of the sensor to the context of the natural language search query, an information gain effected by one or more sensor classes including at least the sensor class to which the sensor belongs;

selecting a subset of the one or more sensor classes consisting of the one or more sensor classes for which the determined information gain exceeds a threshold;

aggregating the one or more sensor classes for which the determined information gain exceeds the threshold with previously determined sensor classes having the information gain that exceeds the threshold, wherein a rating of the one or more sensor classes and the previously determined sensor classes within the aggregated sensor classes corresponds to a number of occurrences of the one or more sensor classes and the previously determined sensor classes in the aggregated sensor classes; and

selecting a highest rated sensor class from among the aggregated sensor classes as the sensor;

determine a second response to the augmented search query, the second response having a second confidence level;

store information to the table of training data that associates the at least one sensor with a value of a change between the first confidence level and the second confidence level; and

augment a subsequent natural language search query according to the stored information.

2. The computer program product of claim 1, wherein the data received from the at least one sensor that is a member of the sensor class validates a determined context of the natural language search query.

3. The computer program product of claim 1, wherein the data received from the at least one sensor that is a member of the sensor class augments a determined context of the natural language search query with a context of the data received from the at least one sensor.

4. The computer program product of claim 3, wherein the table of training data indicates a relevance of the sensor class and the at least one sensor to a context of the natural language search query.

5. The computer program product of claim 1, wherein the at least one sensor is configured to operate as a part of an Internet of things.

6. The computer program product of claim 1, wherein the computer program product is provided in a cloud environment as software as a service to a user who generates the search query.

7. A computer-implemented method, comprising:

determining a preliminary confidence level for a preliminary response to a search query;

determining a sensor having a potential relationship to a context of the search query according to a table of training data by;

determining, based on the table of training data relating relevance of the sensor to the context of the search query, an information gain effected by one or more sensor classes including at least a sensor class to which the sensor belongs;

selecting a subset of the one or more sensor classes consisting of the one or more sensor classes for which the determined information gain exceeds a threshold;

aggregating the one or more sensor classes for which the determined information gain exceeds the threshold with previously determined sensor classes having the information gain that exceeds the threshold, wherein a rating of the one or more sensor classes and the previously determined sensor classes within the aggregated sensor classes corresponds to a number of occurrences of the one or more sensor classes and the previously determined sensor classes in the aggregated sensor classes; and

selecting a highest rated sensor class from among the aggregated sensor classes as the sensor having the potential relationship to the context of the search query;

determining an enhanced search query based on the search query and the sensor;

determining a confidence level for a response to the enhanced search query;

determining that the confidence level varies from the preliminary confidence level;

updating the training data associating the sensor to relevant search query contexts to correspond to the variation in the confidence level from the preliminary confidence level; and

using the updated training data to form a second enhanced search query.

8. The computer-implemented method of claim 7, wherein updating the training data comprises writing data associating the sensor to the relevant search query contexts to correspond to the variation in the confidence level from the preliminary confidence level to a table of sensor relevance training data stored on a non-transitory computer-readable storage medium.

9. The computer-implemented method of claim 7, wherein the table of training data comprises an association between the sensor and a context of the search query.

10. The computer-implemented method of claim 9, wherein the sensor validates the context of the search query.

11. The computer-implemented method of claim 7, wherein the sensor adds additional sensor-based context to the context of the search query.

12. The computer-implemented method of claim 7, wherein the table of training data is determined according to an iterative process of refinement of sensor classes that are determined according to a computed information gain to have a relevance to the context of the search query.

13. The computer-implemented method of claim 7, wherein the computer-implemented method is implemented as a service in a cloud-computing environment.

14. The computer-implemented method of claim 7, wherein the computer-implemented method is implemented as an iterative loop.

15. A system capable of learning, comprising a processor configured to:

parse a search query received from a query source;

determine a preliminary response to the parsed search query, the preliminary response having a preliminary confidence level;

determine at least one sensor class comprising a sensor having a potential relevance to a context of the parsed search query by;

determining, based on a table of training data relating relevance of the sensor to the context of the search query, an information gain effected by one or more sensor classes including at least a sensor class to which the sensor belongs;

selecting a subset of the one or more sensor classes consisting of the one or more sensor classes for which the determined information gain exceeds a threshold;

aggregating the one or more sensor classes for which the determined information gain exceeds the threshold with previously determined sensor classes having the information gain that exceeds the threshold, wherein a rating of the one or more sensor classes and the previously determined sensor classes within the aggregated sensor classes corresponds to a number of occurrences of the one or more sensor classes and the previously determined sensor classes in the aggregated sensor classes; and

selecting a highest rated sensor class from among the aggregated sensor classes as the sensor;

augment the parsed search query with context information derived from the sensor potentially relevant to the context of the parsed search query to form an augmented search query;

determine a response to the augmented search query, the response to the augmented search query having an augmented query confidence level;

determine a change between the preliminary confidence level and the augmented query confidence level;

store training data corresponding to the change between the preliminary confidence level and the augmented query confidence level to the table of training data stored on a non-transitory computer-readable storage medium; and

augment a subsequent search query according to the stored training data.

16. The system of claim 15, wherein the training data is stored to the table of training data on the non-transitory computer-readable storage medium to associate the sensor having the potential relevance to the context of the parsed search query to an actual relevance to the context of the parsed search query, the actual relevance determined according to the change between the preliminary confidence level and the augmented query confidence level.

17. The system of claim 15, wherein the context information derived from the sensor validates the context of the parsed search query.

18. The system of claim 15, wherein the context information derived from the sensor provides additional context to the parsed search query.

19. The system of claim 15, wherein the training data is determined according to an iterative process of determining a change in search query confidence level effected by an addition of sensor data to the search query.