Context-sensitive generation of conversational responses

• US 10,091,140 B2
• Filed: 05/31/2015
• Issued: 10/02/2018
• Est. Priority Date: 05/31/2015
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Galley, Michel, Sordoni, Alessandro, Brockett, Christopher John, Gao, Jianfeng, Dolan, William Brennan, Ji, Yangfeng, Auli, Michael, Mitchell, Margaret Ann, Nie, Jian-Yun
• Primary Examiner(s): Sall, El Hadji
• Application Number:
  US14/726,562
• Publication Number:
  US 20160352656A1
• Time in Patent Office: 1,220 Days
• Field of Search: 709206, 709200
• CPC Class Codes: G06F 40/56 Natural language generation
• CPC Class Codes: G06N 3/044 Recurrent networks, e.g. Hopfield networks
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: H04L 51/02 using automatic reactions or user delegation, e.g. automatic replies or chatbot-generated messages

Abstract

Examples are generally directed towards context-sensitive generation of conversational responses. Context-message-response n-tuples are extracted from at least one source of conversational data to generate a set of training context-message-response n-tuples. A response generation engine is trained on the set of training context-message-response n-tuples. The trained response generation engine automatically generates a context-sensitive response based on a user generated input message and conversational context data. A digital assistant utilizes the trained response generation engine to generate context-sensitive, natural language responses that are pertinent to user queries.

Claims

1. A computer-implemented method for state-free generation of context-sensitive conversational responses, said method comprising:

receiving, by a processor, a message generated by a user, as part of a current conversation with the user;

deriving, by a context derivation component implemented on the processor, context data from a set of context data sources associated with the user, wherein the context data is derived based on its correspondence with the user generated message and the current conversation; and

generating a response to the user generated message, based on the user generated message and the context data, by a response generation engine implemented on the processor, wherein the response generation engine is trained on a plurality of context-message-response n-tuples extracted from at least one source of conversational data such that a response in the at least one context-message-response tuple corresponds to the generated response to the user generated message.

2. The computer-implemented method of claim 1, further comprising:

selecting context-message-response triples from a plurality of context-message-response triples corresponding to a selected context-message data pair, wherein a context-message-response triple comprises a human-generated message, a conversational context, and a reference response corresponding to the user generated message; and

training the response generation engine in a context-sensitive manner using the selected context-message-response triples to form a trained response generation engine.

3. The computer-implemented method of claim 1, further comprising:

extracting context-message-response n-tuples from at least one social media source in a context-sensitive manner, wherein the context-message-response n-tuples are identified for extraction based on a selected context-message data pair, wherein the at least one social media source provides the conversational data in at least one format, wherein a format of the conversational data comprises at least one of a text format, an audio format, or a visual format.

4. The computer-implemented method of claim 1, wherein generating the context-sensitive response to the user generated message further comprises:

generating the context-sensitive response to the user generated message in real-time during the current conversation.

5. The computer-implemented method of claim 1, wherein the context data comprises linguistic context data, the linguistic context data comprising message and response data pairs preceding the user generated message within the current conversation.

6. The computer-implemented method of claim 1, wherein the context data includes non-linguistic context data, and wherein accessing the context data further comprises:

deriving non-linguistic context data from a set of sensors associated with the user in real-time.

7. The computer-implemented method of claim 1, wherein the context data includes non-linguistic context data, and wherein accessing the context data further comprises:

accessing the non-linguistic context data from a set of user data, the non-linguistic context data comprising at least one of preferences data, contacts data, health data, interest data, user activity data, calendar data, activities data, work data, hobby data, or daily routine data.

8. The computer-implemented method of claim 1, wherein the response generation engine comprises a machine learning model.

9. The computer-implemented method of claim 8, wherein the machine learning model includes a neural network.

10. A system for state-free generation of context-sensitive conversational responses, said system comprising:

a processor configured to;

receive a message generated by a user, as part of a current conversation with the user;

derive, by a context derivation component implemented on the processor, context data from a set of context data sources associated with the user, wherein the context data is derived based on its correspondence with the user generated message and the current conversation; and

generate a response to the user generated message, based on the user generated message and the context data, by a response generation engine implemented on the processor, wherein the response generation engine is trained on a plurality of context-message-response n-tuples extracted from at least one source of conversational data such that the response in the at least one context-message-response tuple corresponds to the generated response to the user generated message.

11. The system of claim 10, wherein the processor is further configured to:

select context-message-response triples from a plurality of context-message-response triples corresponding to a selected context-message data pair, wherein a context-message-response triple comprises a human-generated message, a conversational context, and a reference response corresponding to the user generated message; and

train the response generation engine in a context-sensitive manner using the selected context-message-response triples to form a trained response generation engine.

12. The system of claim 10, wherein the processor is further configured to:

extract context-message-response n-tuples from at least one social media source in a context-sensitive manner, wherein the context-message-response n-tuples are identified for extraction based on a selected context-message data pair, wherein the at least one social media source provides the conversational data in at least one format, wherein a format of the conversational data comprises at least one of a text format, an audio format, or a visual format.

13. The system of claim 10, wherein the processor configured to generate the context-sensitive response to the user generated message is further configured to:

generate the context-sensitive response to the user generated message in real-time during the current conversation.

14. The system of claim 10, wherein the context data comprises linguistic context data, the linguistic context data comprising message and response data pairs preceding the user generated message within the current conversation.

15. The system of claim 10, wherein the context data includes non-linguistic context data, and wherein accessing the context data further comprises:

deriving non-linguistic context data from a set of sensors associated with the user in real-time.

16. The system of claim 10, wherein the context data includes non-linguistic context data, and wherein accessing the context data further comprises:

accessing the non-linguistic context data from a set of user data, the non-linguistic context data comprising at least one of preferences data, contacts data, health data, interest data, user activity data, calendar data, activities data, work data, hobby data, or daily routine data.

17. The system of claim 10, wherein the response generation engine comprises a machine learning model.

18. The system of claim 10, wherein the machine learning model includes a neural network.

19. A device comprising:

a processor configured to;

receive a message generated by a user, as part of a current conversation with the user;

derive, by a context derivation component implemented on the processor, context data from a set of context data sources associated with the user, wherein the context data is derived based on its correspondence with the user generated message and the current conversation; and

generate a response to the user generated message, based on the user generated message and the context data, by a response generation engine implemented on the processor, wherein the response generation engine is trained on a plurality of context-message-response n-tuples extracted from at least one source of conversational data-such that the response in the at least one context-message-response tuple corresponds to the generated response to the user generated message.

20. The device of claim 19, wherein the processor is further configured to:

select context-message-response triples from a plurality of context-message-response triples corresponding to a selected context-message data pair, wherein a context-message-response triple comprises a human-generated message, a conversational context, and a reference response corresponding to the user generated message; and

train the response generation engine in a context-sensitive manner using the selected context-message-response triples to form a trained response generation engine.