Natural language intent and location determination method and apparatus

• US 10,394,954 B2
• Filed: 02/27/2017
• Issued: 08/27/2019
• Est. Priority Date: 02/27/2017
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Vainas, Oded, Pereg, Oren, Mendels, Omri, Soffer, Ronen
• Primary Examiner(s): Cyr, Leonard Saint
• Application Number:
  US15/443,043
• Publication Number:
  US 20180246877A1
• Time in Patent Office: 911 Days
• Field of Search: 704 7- 10, 704275
• CPC Class Codes: G06F 40/295 Named entity recognition
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

Methods, apparatus, and system to parse an unstructured, natural language input of a user, infer a semantic meaning of an intent of the user, and determine a present or future user state, including a location and a time, in which the intent can be fulfilled.

Claims

1. An apparatus for computing to infer a state of a user in which an intent of the user can be performed, comprising:

a computer processor, a memory, and a natural language processing hardware accelerator with;

an intent extraction module to determine the intent of the user from a natural language input;

a state of a user module to determine a present or predicted state of the user, wherein the state of the user module is to apply a machine learning classifier module comprising a supervised learning model trained on one or both of a sensor information and a computer application information on the apparatus;

an intent-to-state matching module to match the intent with the present or predicted state of the user; and

an interface to receive the natural language input or a binary conversion of the natural language input and to output a notice regarding the match between the present or predicted state of the user and the intent of the user;

wherein

to match the intent with the present or predicted state of the user, the intent-to-state matching module is to predict a state of the user the intent can be performed in, wherein to predict the state of the user the intent can be performed in comprises to determine a present or future location of the user according to a route of the user and a location category of such present or future location of the user, obtain a location of the intent and a category of the location of the intent, and match the location category of such present or future location of the user to the category of the location of the intent.

2. The apparatus according to claim 1, wherein to determine the intent of the user from the natural language input, the intent extraction module is to apply a machine learning classifier to convert the natural language input into structured output arguments, wherein the structured output arguments comprise a verbal predicate comprising the intent of the user, a theme, and a semantic representation of the theme, wherein the theme and/or semantic representation comprises at least one object of the intent, a location of the intent, and a day-time of the intent.

3. The apparatus according to claim 2, wherein to convert the natural language input into the structured output arguments comprises to extract terms from the natural language input and determine a likely sense of each term with the machine learning classifier according to a vector space representation of the terms, wherein the vector space representation of the terms comprises a distributional property of the terms in a sample of natural language.

4. The apparatus according to claim 1, wherein the intent extraction module is further to infer at least one of a time, place, state of the user, or a location category of the location of the intent, based on application of a knowledge graph or trained topic model trained on past intents and reminders set by multiple users with respect to the past intents.

5. The apparatus according to claim 1, wherein to determine the present or predicted state of the user, the state of the user module is to estimate a present or future location of the user, a category of the present or future location of the user, a present or future action of the user, or a present or future availability of the user, wherein the state of the user module is further to form a route of the user based on a time series of the present location of the user and on at least one of routine analysis, visitation analysis, or mobility pattern analysis of the time series, and wherein the state of the user module is further to predict the future location of the user based the route of the user and a schedule of the user.

6. The apparatus according to claim 1, wherein the machine learning classifier module further comprises a learning algorithm to analyze the natural language input according to one or both of classification or regression analysis.

7. The apparatus according to claim 1, wherein to output the notice regarding the matched present state of the user or the predicted the state of the user, the interface is to output at least one of a reminder to engage in the intent, a confirmation of the intent of the user, or a confirmation of the match between the present or predicted state of the user and the intent of the user.

8. A computer implemented method, wherein the method comprises, with a natural language processing hardware accelerator:

receiving a natural language input or a binary conversion of the natural language input;

determining an intent of a user from a natural language input;

determining a present or predicted state of the user with a machine learning classifier comprising a supervised learning model trained on one or both of a sensor information and a computer application information;

matching the intent with the present or predicted state of the user; and

outputting a notice regarding the match between the present or predicted state of the user and the intent of the user;

wherein matching the intent with the present or predicted state of the user comprises predicting a state of the user the intent can be performed in and wherein predicting a state of the user the intent can be performed in comprises determining a present or future location of the user according to a route of the user and a location category of such present or future location of the user, obtaining a location of the intent and a category of the location of the intent, and matching the location category of such present or future location of the user to the category of the location of the intent.

9. The method according to claim 8, wherein determining the intent of the user from the natural language input comprises converting the natural language input into structured output arguments using a machine learning classifier, wherein the structured output arguments comprise a verbal predicate comprising the intent of the user, a theme, and a semantic representation of the theme, wherein the theme and/or semantic representation comprises at least one object of the intent, a location of the intent, and a day-time of the intent.

10. The method according to claim 9, wherein converting the natural language input into the structured output arguments comprises extracting terms from the natural language input and determining a likely sense of each term with the machine learning classifier according to a vector space representation of the terms, wherein the vector space representation of the terms comprises a distributional property of the terms in a sample of natural language.

11. The method according to claim 8, further comprising inferring at least one of a time, place, state of the user, or a location category of the location of the intent based on applying a knowledge graph or trained topic model trained on past intents and reminders set by multiple users with respect to the past intents.

12. The method according to claim 8, wherein determining the present or predicted state of the user comprises estimating a present or future location of the user, a category of the present or future location of the user, a present or future action of the user, or a present or future availability of the user;

further comprising determining the present location of the user from a location information of the user from a location service;

further comprising predicting the future location of the user based a route of the user and a schedule of the user; and

further comprising determining the category of the present or future location of the user according to past intents and reminders set with respect to the present or future location of the user.

13. The method according to claim 8, wherein the machine learning classifier comprises a learning algorithm analyzing the natural language input through one or both of classification or regression analysis.

14. An apparatus for computing, comprising:

means to receive a natural language input or a binary conversion of the natural language input;

means to determine an intent of a user from a natural language input;

means to determine a present or predicted state of the user;

means to match the intent with the present or predicted state of the user;

means to output a notice regarding the match between the present or predicted state of the user and the intent of the user;

wherein

means to determine a present or predicted state of the user comprises means to apply a machine learning classifier comprising a supervised learning model trained on one or both of a sensor information and a computer application information on the apparatus; and

means to match the intent with the present or predicted state of the user comprises means to predict a state of the user the intent can be performed in, wherein means to predict the state of the user the intent can be performed in comprises means to determine a present or future location of the user according to a route of the user and a location category of such present or future location of the user, means to obtain a location of the intent and a category of the location of the intent, and means to match the location category of such present or future location of the user to the category of the location of the intent.

15. The apparatus according to claim 14, wherein means to determine the intent of the user from the natural language input comprises means to apply a machine learning classifier to convert the natural language input into structured output arguments, wherein the structured output arguments comprise a verbal predicate comprising the intent of the user, a theme, and a semantic representation of the theme, wherein the theme and/or semantic representation of the theme comprises at least one object of the intent, a location of the intent, and a day-time of the intent, wherein the machine-learning classifier comprises a supervised learning model and a learning algorithm to analyze the natural language input for one or both of classification or regression analysis, and wherein the supervised learning model comprises at least one of a support vector machine, a logistic regression, a decision tree, an ensemble model, a k-nearest neighbors model, or a neural network.

16. The apparatus according to claim 15, wherein means to convert the natural language input into the structured output arguments comprises means to extract terms from the natural language input and determine a likely sense of each term with the machine learning classifier according to a vector space representation of the terms, wherein the vector space representation of the terms comprises a distributional property of the terms in a sample of natural language.

17. The apparatus according to claim 14, further comprising means to infer at least one of a time, place, state of the user, or a location category of the location of the intent using means to apply a knowledge graph or trained topic model trained on past intents and reminders set by multiple users with respect to the past intents.

18. The apparatus according to claim 14, wherein means to determine the present or predicted state of the user comprises mean to estimate a present or future location of the user, a category of the present or future location of the user, a present or future action of the user, or a present or future availability of the user, wherein the means to determine the present location of the user comprises means to receive a location information of the user from a location service, means to cluster the location information of the user into a cluster;

further comprising means to form a route of the user based on a time series of the present location of the user and means to predict the future location of the user based the route of the user and a schedule of the user.

19. The apparatus according to claim 14, wherein the machine learning classifier comprises means to analyze the natural language input through one or both of classification or regression analysis.

20. The apparatus according to claim 14, wherein means to output the notice regarding the matched present state of the user or the predicted the state of the user comprises means to output at least one of a reminder to engage in the intent, a confirmation of the intent of the user, or a confirmation of the match between the present or predicted state of the user and the intent of the user.

21. One or more non-transitory computer-readable media comprising instructions that cause a computer device, in response to execution of the instructions by a processor of the computer device, to:

receive a natural language input or a binary conversion of the natural language input;

determine an intent of a user from a natural language input;

determine a present or predicted state of the user with a machine learning classifier comprising a supervised learning model trained on one or both of a sensor information and a computer application information of the computer device;

match the intent with the present or predicted state of the user; and

output a notice regarding the match between the present or predicted state of the user and the intent of the user;

wherein match the intent with the present or predicted state of the user comprises predict a state of the user the intent can be performed in and wherein predict a state of the user the intent can be performed in comprises determine a present or future location of the user according to a route of the user and a location category of such present or future location of the user, obtain a location of the intent and a category of the location of the intent, and match the location category of such present or future location of the user to the category of the location of the intent.

22. The non-transitory computer-readable media according to claim 21, wherein determine the intent of the user from the natural language input comprises apply a machine learning classifier to convert the natural language input into structured output arguments, wherein the structured output arguments comprise a verbal predicate comprising the intent of the user, a theme, and a semantic representation of the theme, wherein the theme and/or semantic representation comprises at least one object of the intent, a location of the intent, and a day-time of the intent.

23. The non-transitory computer-readable media according to claim 22, wherein convert the natural language input into the structured output arguments comprises extract terms from the natural language input and determine a likely sense of each term with the machine learning classifier according to a vector space representation of the terms, wherein the vector space representation of the terms comprises a distributional property of the terms in a sample of natural language.

24. The non-transitory computer-readable media according to claim 21, wherein the instructions are further to cause to the processor to infer at least one of a time, place, state of the user, or a location category of the location of the intent.

25. The non-transitory computer-readable media according to claim 21, wherein the machine learning classifier further comprises a learning algorithm to analyze the natural language input for one or both of classification or regression analysis.