Granted Patent B2

US 12,406,668 · App. 18/213,525 · Granted Sep 2, 2025

Inventors: Xiao Yan Lu (Bellevue, WA); Amir Kantor (Haifa, IL); Ido Priness (Kiryat Ono, IL); Shiraz Jitendra Cupala (Snohomish, WA); Kevin Michael Carter (Atlanta, GA); Adi Miller (Ramat Hasharon, IL); Kumud Ranjan (Redmond, WA); Shyam Gupta (Surrey, CA); Gautam Jain (Surrey, CA); Yasemin Cenberoglu (Montreal, CA); Shai Ifrach (Yavne, IL); Shlomi Maliah (Rosh Hayin, IL); Gilad Gildin (Tel Aviv, IL); Ofek David (Tel-Aviv, IL); Eleonora Shtotland (Herzliya, IL); Jaime Teevan (Bellevue, WA); Matthew Jonathan Gardner (Irvine, CA); Lan Ye (Issaquah, WA)

Assignee: Microsoft Technology Licensing, LLC

G10L15/22G10L15/063G10L2015/221G10L2015/223

Abstract

Claims (91)

1. A system providing a personalized assistant within a network-based communication session, the system comprising:

a processor; and

a memory storage device storing instructions thereon, which, when executed by the processor, cause the system to perform operations comprising:

determining a first communication session participant has joined a network-based communication session after a threshold duration of time subsequent to the start time of the network-based communication session; and

responsive to determining the first communication session participant has joined the network-based communication session after the threshold duration of time:

obtaining content associated with the network-based communication session, the content originating during a window of time spanning the start time of the network-based communication session and the time at which the communication session participant joined the network-based communication session;

creating request data for a pre-trained generative language model based upon a portion of the content associated with the network-based communication session and an instruction requesting, as output, a predetermined number of suggested utterances not present in the portion of the content, each utterance relating to one or more topics corresponding to the portion of the content;

transforming the request data to a command based upon a command template;

providing the command to a pre-trained generative language model with the request data;

receiving a response from the generative language model, the response including the predetermined number of suggested utterances; and

causing one or more of the predetermined number of suggested utterances to be presented to the communication session participant in a graphical user interface of the network-based communication session, while the network-based communication session is in session.

2. The system of claim 1 , wherein the instruction requesting, as output, the predetermined number of suggested utterances not present in the portion of the content includes a request to generate questions that have not already been asked by another communication session participant, or a request to generate questions that exclude questions that have already been asked by another communication session participant.

3. The system of claim 1 , wherein the content associated with the network-based communication session is structured as a communication session transcript having a plurality of chronologically ordered content items, wherein each content item in the plurality of chronologically ordered content items represents a communication made by a communication session participant and includes data indicating the name of the communication session participant who made the communication, wherein the instruction to generate the predetermined number of suggested utterances not present in the portion of the content includes a request to include the name of a communication session participant to whom a question should be directed.

4. The system of claim 1 , wherein the memory storage device is storing instructions, which, when executed by the one or more processors, cause the system to perform additional operations comprising:

providing as a second input to the pre-trained generative language model an instruction to generate a summary description of the network-based communication session, based on the content;

receiving as output from the generative language model a summary description of the network-based communication session; and

causing the summary description of the network-based communication session to be presented to the communication session participant in a graphical user interface of the network-based communication session.

5. The system of claim 1 , wherein the memory storage device is storing instructions, which, when executed by the one or more processors, cause the system to perform additional operations comprising:

segmenting the content into a plurality of segments, each segment in the plurality of segments having a size that is based on a maximum input size requirement of the pre-trained generative language model;

using the generative language model to generate a summary description of the network-based communication session, by:

for each segment of the plurality of segments, providing as input to the pre-trained generative language model content from the segment, and ii) an instruction to generate a summary description of the network-based communication session, based on the content;

receiving as output from the generative language model a summary description of the network-based communication session, for each segment of the plurality of segments;

providing to the pre-trained generative language model a final input, the final input including i) the summary description of the network-based communication session, as output by the generative language model for each segment of the plurality of segments and ii) an instruction to generate an overall summary description of the network-based communication session, based on the summary description of the network-based communication session as output by the generative language model for each segment of the plurality of segments;

responsive to providing the final input to the generative language model, receiving as output from the generative language model an overall summary description of the network-based communication session, based on the summary description of the network-based communication session as output by the generative language model for each segment of the plurality of segments; and

causing the overall summary description of the network-based communication session to be presented to the communication session participant in a graphical user interface of the network-based communication session.

6. The system of claim 1 , wherein the memory storage device is storing instructions, which, when executed by the one or more processors, cause the system to perform additional operations comprising:

segmenting the content into a plurality of segments, each segment in the plurality of segments having a size that is based on a maximum input size requirement of the pre-trained generative language model;

using the generative language model to generate a summary description of the network-based communication session, by:

for a first segment in the plurality of segments, providing as input to the pre-trained generative language model i) content from the first segment, and ii) an instruction to generate a summary description of the network-based communication session, based on the content from the first segment;

for each segment in the plurality of segments subsequent to the first segment, providing as input to the pre-trained generative language model i) the summary description of the network-based communication session output by the pre-trained generative language model based on a prior segment and content from the segment, and ii) an instruction to generate a summary description of the network-based communication session;

receiving as output from the generative language model a final summary description of the network-based communication session, based on the generative language model processing a final prompt for a last segment in the plurality of segments; and

causing the final summary description of the network-based communication session to be presented to the communication session participant in a graphical user interface of the network-based communication session.

7. The system of claim 1 , wherein the content associated with the in-session network-based communication session is structured as a communication session transcript having a plurality of chronologically ordered content items, wherein each content item in the plurality of chronologically ordered content items represents a communication made by a communication session participant and includes data indicating the name of the communication session participant who made the communication, wherein the memory storage device is storing instructions, which, when executed by the one or more processors, cause the system to perform additional operations comprising:

determine existence of a specific type of relationship between the first communication session participant and a second communication session participant;

extracting from the content one or more content items representing a communication made by the second communication session participant;

providing as input to a pre-trained generative language model, i) the one or more extracted content items, and ii) an instruction to generate a summary description of communications made by the second communication session participant;

receiving as output from the generative language model the summary description of communications made by the second communication session participant; and

causing the summary description of communications made by the second communication session participant to be presented to the first communication session participant in a graphical user interface of the network-based communication session.

8. The system of claim 1 , wherein the pre-trained generative language model has been fine-tuned, using a supervised learning technique, to generate suggested utterances not present in a portion of content based on a conversation of communication session participants as expressed in content, wherein a training dataset used in fine-tuning the pre-trained generative language model includes a plurality of instances of a communication session transcript having a plurality of chronologically ordered content items, each content item in the plurality of chronologically ordered content items representing a communication made by a communication session participant, wherein one or more communications have been labeled as questions, statements or opinions.

9. A method for providing a personalized assistant within a network-based communication session, the method comprising:

determining a first communication session participant has joined a network-based communication session after a threshold duration of time subsequent to the start time of the network-based communication session; and

responsive to determining the first communication session participant has joined the network-based communication session after the threshold duration of time:

obtaining content associated with the network-based communication session, the content originating during a window of time spanning the start time of the network-based communication session and the time at which the communication session participant joined the network-based communication session;

creating request data for a pre-trained generative language model based upon a portion of the content associated with the network-based communication session and an instruction requesting, as output, a predetermined number of suggested utterances not present in the portion of the content, each utterance relating to one or more topics corresponding to the portion of the content;

transforming the request data to a command based upon a command template;

providing the command to a pre-trained generative language model with the request data;

receiving a response from the generative language model, the response including the predetermined number of suggested utterances; and

causing one or more of the predetermined number of suggested utterances to be presented to the communication session participant in a graphical user interface of the network-based communication session, while the network-based communication session is in session.

10. The method of claim 9 , wherein the instruction requesting, as output, the predetermined number of suggested utterances not present in the portion of the content includes a request to generate questions that have not already been asked by another communication session participant, or a request to generate questions that exclude questions that have already been asked by another communication session participant.

11. The method of claim 9 , wherein the content associated with the network-based communication session is structured as a communication session transcript having a plurality of chronologically ordered content items, wherein each content item in the plurality of chronologically ordered content items represents a communication made by a communication session participant and includes data indicating the name of the communication session participant who made the communication, wherein the instruction to generate the predetermined number of suggested utterances not present in the portion of the content includes a request to include the name of a communication session participant to whom a question should be directed.

12. The method of claim 9 , further comprising:

providing as a second input to the pre-trained generative language model an instruction to generate a summary description of the network-based communication session, based on the content;

receiving as output from the generative language model a summary description of the network-based communication session; and

causing the summary description of the network-based communication session to be presented to the communication session participant in a graphical user interface of the network-based communication session.

13. The method of claim 9 , wherein the memory storage device is storing instructions, which, when executed by the one or more processors, cause the system to perform additional operations comprising:

segmenting the content into a plurality of segments, each segment in the plurality of segments having a size that is based on a maximum input size requirement of the pre-trained generative language model;

using the generative language model to generate a summary description of the network-based communication session, by:

for each segment of the plurality of segments, providing as input to the pre-trained generative language model content from the segment, and ii) an instruction to generate a summary description of the network-based communication session, based on the content;

receiving as output from the generative language model a summary description of the network-based communication session, for each segment of the plurality of segments;

providing to the pre-trained generative language model a final input, the final input including i) the summary description of the network-based communication session, as output by the generative language model for each segment of the plurality of segments and ii) an instruction to generate an overall summary description of the network-based communication session, based on the summary description of the network-based communication session as output by the generative language model for each segment of the plurality of segments;

responsive to providing the final input to the generative language model, receiving as output from the generative language model an overall summary description of the network-based communication session, based on the summary description of the network-based communication session as output by the generative language model for each segment of the plurality of segments; and

causing the overall summary description of the network-based communication session to be presented to the communication session participant in a graphical user interface of the network-based communication session.

14. The system of claim 9 , further comprising:

segmenting the content into a plurality of segments, each segment in the plurality of segments having a size that is based on a maximum input size requirement of the pre-trained generative language model:

using the generative language model to generate a summary description of the network-based communication session, by:

for a first segment in the plurality of segments, providing as input to the pre-trained generative language model i) content from the first segment, and ii) an instruction to generate a summary description of the network-based communication session, based on the content from the first segment;

for each segment in the plurality of segments subsequent to the first segment, providing as input to the pre-trained generative language model i) the summary description of the network-based communication session output by the pre-trained generative language model based on a prior segment and content from the segment, and ii) an instruction to generate a summary description of the network-based communication session;

receiving as output from the generative language model a final summary description of the network-based communication session, based on the generative language model processing a final prompt for a last segment in the plurality of segments; and

causing the final summary description of the network-based communication session to be presented to the communication session participant in a graphical user interface of the network-based communication session.

15. The method of claim 9 , wherein the content associated with the in-session network-based communication session is structured as a communication session transcript having a plurality of chronologically ordered content items, wherein each content item in the plurality of chronologically ordered content items represents a communication made by a communication session participant and includes data indicating the name of the communication session participant who made the communication, wherein the method further comprises:

determining existence of a specific type of relationship between the first communication session participant and a second communication session participant;

extracting from the content one or more content items representing a communication made by the second communication session participant;

providing as input to a pre-trained generative language model, i) the one or more extracted content items, and ii) an instruction to generate a summary description of communications made by the second communication session participant;

receiving as output from the generative language model the summary description of communications made by the second communication session participant; and

causing the summary description of communications made by the second communication session participant to be presented to the first communication session participant in a graphical user interface of the network-based communication session.

16. The method of claim 9 , wherein the pre-trained generative language model has been fine-tuned, using a supervised learning technique, to generate suggested utterances not present in a portion of content based on a conversation of communication session participants as expressed in content, wherein a training dataset used in fine-tuning the pre-trained generative language model includes a plurality of instances of a communication session transcript having a plurality of chronologically ordered content items, each content item in the plurality of chronologically ordered content items representing a communication made by a communication session participant, wherein one or more communications have been labeled as questions, statements or opinions.

17. A system providing a personalized assistant within a network-based communication session, the system comprising:

means for determining a first communication session participant has joined a network-based communication session after a threshold duration of time subsequent to the start time of the network-based communication session; and

responsive to determining the first communication session participant has joined the network-based communication session after the threshold duration of time:

means for obtaining content associated with the network-based communication session, the content originating during a window of time spanning the start time of the network-based communication session and the time at which the communication session participant joined the network-based communication session;

means for creating request data for a pre-trained generative language model based upon a portion of the content associated with the network-based communication session and an instruction requesting, as output, a predetermined number of suggested utterances not present in the portion of the content, each utterance relating to one or more topics corresponding to the portion of the content;

means for transforming the request data to a command based upon a command template;

means for providing the command to a pre-trained generative language model with the request data;

means for receiving a response from the generative language model, the response including the predetermined number of suggested utterances; and

means for causing one or more of the predetermined number of suggested utterances to be presented to the communication session participant in a graphical user interface of the network-based communication session, while the network-based communication session is in session.

18. The system of claim 17 , wherein the instruction requesting, as output, the predetermined number of suggested utterances not present in the portion of the content includes a request to generate questions that have not already been asked by another communication session participant, or a request to generate questions that exclude questions that have already been asked by another communication session participant.

19. The system of claim 17 , wherein the content associated with the network-based communication session is structured as a communication session transcript having a plurality of chronologically ordered content items, wherein each content item in the plurality of chronologically ordered content items represents a communication made by a communication session participant and includes data indicating the name of the communication session participant who made the communication, wherein the instruction to generate the predetermined number of suggested utterances not present in the portion of the content includes a request to include the name of a communication session participant to whom a question should be directed.

20. The system of claim 17 , further comprising:

means for providing as a second input to the pre-trained generative language model an instruction to generate a summary description of the network-based communication session, based on the content;

means for receiving as output from the generative language model a summary description of the network-based communication session; and

means for causing the summary description of the network-based communication session to be presented to the communication session participant in a graphical user interface of the network-based communication session.

Assignments (1)

Continuity (2)

References Cited (73)