Granted Patent B2

US 12,271,908 · App. 17/370,092 · Granted Apr 8, 2025

Inventors: Radu Marinescu (Dublin, IE); Akihiro Kishimoto (Setagaya, JP); Yufang Hou (Dublin, IE)

Assignee: International Business Machines Corporation

G06Q30/016G06F40/205G06F40/279G06N20/00G06Q10/06316G06Q10/0633G06Q10/103G06F16/23

Abstract

Claims (93)

1. A computer-implemented method for an automated information technology (“IT”) support system, the computer-implemented method comprising:

receiving, by one or more processors, a set of sentences extracted from a natural language conversation occurring between an IT support system and a user through a channel used to log an IT support ticket;

extracting, by the one or more processors, an initial state and a goal state from the set of sentences using a Natural Language Classifier, wherein-the Natural Language Classifier evaluates the set of sentences against a plurality of support vector machines and a convolutional neural network based on a first deep learning model;

extracting, by the one or more processors, one or more actions from the set of sentences;

converting, by the one or more processors, the one or more actions into one or more into corresponding planning design description language (“PDDL”) text formal actions;

creating, by the one or more processors, a formal planning model incorporating the extracted initial state, the goal state, and the one or more formal actions, based on the converted PDDL text;

determining, by the one or more processors, whether at least one of the one or more formal actions are not complete based on a first machine learning model;

responsive to determining the at least one of the one or more formal actions are not complete, completing, by the one or more processors, the one or more formal actions with one or more missing parts using the first machine learning model enabled to predict the one or more missing parts of the at least one of the one or more formal actions that are not complete to produce one or more complete formal actions;

generating, by the one or more processors, an executable plan to move from the initial state to the goal state by mapping the one or more complete formal actions to one or more executable scripts, wherein an executable script is a code that can be executed on a computing device; and

executing, by the one or more processors, the one or more executable scripts according to the sequence of the one or more complete formal actions.

2. The computer-implemented method of claim 1 , wherein extracting the initial state and the goal state from the set of sentences using the Natural Language Classifier further comprises:

running, by the one or more processors, a seq 2 seq model that translates a sentence identified as the initial state and a sentence identified as the goal state into a logical expression; and

adding, by the one or more processors, a corresponding logical expression to the sentence identified as the initial state and a corresponding logical expression to the sentence identified as the goal state.

3. The computer-implemented method of claim 1 , wherein extracting the one or more actions from the set of sentences further comprises:

extracting, by the one or more processors, a name of each of the one or more actions from the set of sentences;

extracting, by the one or more processors, a reason for taking each of the one or more actions from the set of sentences;

running, by the one or more processors, a semantic parser on the set of sentences to detect state transitions when the one or more actions from the set of sentences are applied; and

extracting, by the one or more processors, one or more pre-conditions and one or more effects of the one or more actions from the set of sentences.

4. The computer-implemented method of claim 1 , wherein completing the one or more formal actions with the one or more missing parts using the first machine learning model enabled to predict the one or more missing parts of the at least one of the one or more formal actions that are not complete to produce the one or more complete formal actions further comprises:

requesting, by the one or more processors, feedback from the user regarding the one or more formal actions with the one or more missing parts;

enabling, by the one or more processors, the user to review the one or more formal actions with the one or more missing parts;

enabling, by the one or more processors, the user to revise the one or more formal actions with the one or more missing parts manually through an iterative dialog system;

updating, by the one or more processors, the formal planning model with the one or more missing parts; and

updating, by the one or more processors, a database with the one or more missing parts.

5. The computer-implemented method of claim 1 , wherein producing the executable plan to move from the initial state to the goal state by mapping the one or more complete formal actions to the one or more executable scripts further comprises:

mapping, by the one or more processors, the one or more complete formal actions of the executable plan to the one or more executable scripts until all of the one or more complete formal actions of the executable plan have a corresponding executable script; and

producing, by the one or more processors, the executable plan by linking together the one or more complete formal actions in such a way that will satisfy the goal state.

6. The computer-implemented method of claim 5 , wherein mapping the one or more complete formal actions of the executable plan to the one or more executable scripts until all of the one or more complete formal actions of the executable plan have the corresponding executable script further comprises:

selecting, by the one or more processors, the one or more complete formal actions of the executable plan with a missing executable script;

enabling, by the one or more processors, the user to provide the missing executable script; and

updating, by the one or more processors, the database with the missing executable script provided by the user.

7. The computer-implemented method of claim 1 , wherein the first deep learning model is an encoder-decoder deep learning model.

8. The computer-implemented method of claim 1 , wherein the first machine learning model is trained using at least one set of machine-learned data, wherein the machine-learned data is collected from self-learning.

9. A computer program product for an automated information technology (“IT”) support system, the computer program product comprising:

one or more non-transitory computer readable storage media and program instructions stored on the one or more non-transitory computer readable storage media, the program instructions comprising:

program instructions to receive a set of sentences extracted from a natural language conversation occurring between an IT support system and a user through a channel used to log an IT support ticket;

program instructions to extract an initial state and a goal state from the set of sentences using a Natural Language Classifier, wherein the the Natural Language Classifier evaluates the set of sentences against a plurality of support vector machines and a convolutional neural network based on a first deep learning model;

program instructions to extract one or more actions from the set of sentences;

program instructions to convert the one or more actions into corresponding planning design description language (“PDDL”) text formal actions;

program instructions to create a formal planning model incorporating the extracted initial state, the goal state, and the one or more formal actions, based on the converted PDDL text;

program instructions to determine whether at least one of the one or more formal actions are not complete based on a first machine learning model;

responsive to determining the at least one of the one or more formal actions are not complete, program instructions to complete the one or more formal actions with one or more missing parts using the first machine learning model enabled to predict the one or more missing parts of the at least one of the one or more formal actions that are not complete to produce one or more complete formal actions;

program instructions to generate an executable plan to move from the initial state to the goal state by mapping the one or more complete formal actions to one or more executable scripts, wherein an executable script is a code that can be executed on a computing device; and

program instructions to execute the one or more executable scripts according to the sequence of the one or more complete formal actions.

10. The computer program product of claim 9 , wherein the program instructions to extract the one or more actions from the set of sentences further comprises:

program instructions to extract a name of each of the one or more actions from the set of sentences;

program instructions to extract a reason for taking each of the one or more actions from the set of sentences;

program instructions to run a semantic parser on the set of sentences to detect state transitions when the one or more actions are applied; and

program instructions to extract one or more pre-conditions and one or more effects of the one or more actions.

11. The computer program product of claim 9 , wherein the program instructions to complete the one or more formal actions with the one or more missing parts using the first machine learning model enabled to predict the one or more missing parts of the at least one of the one or more formal actions that are not complete to produce the one or more complete formal actions further comprises:

program instructions to request feedback from the user regarding the one or more formal actions with the one or more missing parts;

program instructions to enable the user to review the one or more formal actions with the one or more missing parts;

program instructions to enable the user to revise the one or more formal actions with the one or more missing parts manually through an iterative dialog system;

program instructions to update the formal planning model with the one or more missing parts; and

program instructions to update a database with the one or more missing parts.

12. The computer program product of claim 9 , wherein the program instructions to produce the executable plan to move from the initial state to the goal state by mapping the one or more complete formal actions to the one or more executable scripts further comprises:

program instructions to map the one or more complete formal actions of the executable plan to one or more executable scripts until all of the one or more complete formal actions of the executable plan have a corresponding executable script; and

program instructions to produce the executable plan by linking together the one or more complete formal actions in such a way that will satisfy the goal state.

13. The computer program product of claim 12 , wherein the program instructions to map the one or more complete formal actions of the executable plan to the one or more executable scripts until all of the one or more complete formal actions of the executable plan have the corresponding executable script further comprises:

program instructions to select the one or more complete formal actions of the executable plan with a missing executable script;

program instructions to enable the user to provide the missing executable script; and

program instructions to update the database with the missing executable script provided by the user.

14. A computer system for an automated information technology (“IT”) support system, the computer system comprising:

one or more computer processors;

one or more non-transitory computer readable storage media;

program instructions collectively stored on the one or more non-transitory computer readable storage media for execution by at least one of the one or more computer processors, the stored program instructions comprising:

program instructions to receive a set of sentences extracted from a natural language conversation occurring between an IT support system and a user through a channel used to log an IT support ticket;

program instructions to extract an initial state and a goal state from the set of sentences using a Natural Language Classifier, wherein the Natural Language Classifier evaluates the set of sentences against a plurality of support vector machines and a convolutional neural network using a first deep learning model;

program instructions to extract one or more actions from the set of sentences;

program instructions to convert the one or more actions into corresponding planning design description language (“PDDL”) text formal actions;

program instructions to create a formal planning model incorporating the extracted initial state, the goal state, and the one or more formal actions, based on the converted PDDL text;

program instructions to determine whether at least one of the one or more formal actions are not complete based on a first machine learning model;

responsive to determining the at least one of the one or more formal actions are not complete, program instructions to complete the one or more formal actions with one or more missing parts using the first machine learning model enabled to predict the one or more missing parts of the at least one of the one or more formal actions that are not complete to produce one or more complete formal actions;

program instructions to generate an executable plan to move from the initial state to the goal state by mapping the one or more complete formal actions to one or more executable scripts, wherein an executable script is a code that can be executed on a computing device;

program instructions to execute the one or more executable scripts according to the sequence of the one or more complete formal actions.

15. The computer system of claim 14 , wherein the program instructions to extract the one or more actions from the set of sentences further comprises:

program instructions to extract a name of each of the one or more actions from the set of sentences;

program instructions to extract a reason for taking each of the one or more actions from the set of sentences;

program instructions to run a semantic parser on the set of sentences to detect state transitions when the one or more actions are applied; and

program instructions to extract one or more pre-conditions and one or more effects of the one or more actions.

16. The computer system of claim 14 , wherein the program instructions to complete the one or more formal actions with the one or more missing parts using the first machine learning model enabled to predict the one or more missing parts of the at least one of the one or more formal actions that are not complete to produce the one or more complete formal actions further comprises:

program instructions to request feedback from the user regarding the one or more formal actions with the one or more missing parts;

program instructions to enable the user to review the one or more formal actions with the one or more missing parts;

program instructions to enable the user to revise the one or more formal with the one or more missing parts manually through an iterative dialog system;

program instructions to update the formal planning model with the one or more missing parts; and

program instructions to update a database with the one or more missing parts.

17. The computer system of claim 14 , wherein the program instructions to produce the executable plan to move from the initial state to the goal state by mapping the one or more complete formal actions to the one or more executable scripts further comprises:

program instructions to map the one or more complete formal actions of the executable plan to one or more executable scripts until all of the one or more complete formal actions of the executable plan have a corresponding executable script; and

program instructions to produce the executable plan by linking together the one or more complete formal actions in such a way that will satisfy the goal state.

18. The computer system of claim 17 , wherein the program instructions to map the one or more complete formal actions of the executable plan to the one or more executable scripts until all of the one or more complete formal actions of the executable plan have the corresponding executable script further comprises:

program instructions to select the one or more complete formal actions of the executable plan with a missing executable script;

program instructions to enable the user to provide the missing executable script; and

program instructions to update the database with the missing executable script provided by the user.

Assignments (1)

Continuity (1)

References Cited (35)