Method and system for generative AI with dynamic persona adaptation and applications thereof
The present teaching relates to conduct persona-adaptive communications with a customer at a geo-locale. Transcripts of a current and historic communications involving the customer are used to characterize the persona of the customer. Transcripts of historic communications with customers at the geo-locale are used to characterize the persona of the geo-locale. Current persona of the customer exhibited in the current communication is combined with the customer's persona and the geo-locale's persona to compute a response input vector, A language model generates, based on the response input vector, a persona-adaptive response, which is then sent to the customer a response.
1 . A method, comprising:
receiving a transcript of a current communication with a customer at a geo-locale;
retrieving a first set of transcripts of historic communications associated with the customer and a second set of transcripts of historic communications that occurred at the geo-locale;
computing a user persona vector based on
a hyper contextual persona vector obtained based on the transcript and the first set of transcripts, wherein the hyper contextual persona vector characterizes a first persona of the customer exhibited in the current and the historic communications, and
a geo-localized persona vector obtained based on the second set of transcripts, wherein the geo-localized persona vector characterizes a second persona of the geo-locale;
computing a current persona vector based on the transcript to characterize a current persona of the customer exhibited in the current communication;
generating, by a language model, a persona-adaptive response based on the user persona vector and the current persona vector, wherein the persona-adaptive response is expressed with a persona dynamically adapted with respect to the customer and the geo-locale; and
sending the persona-adaptive response to the customer.
2 . The method of claim 1 , wherein the computing the user persona vector comprises:
processing the transcript and the first set of transcripts to generate a first processed result;
identifying a first plurality of features relating to the first persona of the customer based on the first processed result; and
computing the hyper contextual persona vector based on the first plurality of features and user persona embeddings previously trained via machine learning.
3 . The method of claim 2 , wherein the first plurality of features include at least one of:
user-specific vocabulary with words/phrases used by the customer in the current and historic communications;
user-specific grammar used by the customer in the current and historic communications; and
user-specific demographic information about the customer.
4 . The method of claim 2 , further comprising:
processing the second set of transcripts to generate a second processed result;
determining a second plurality of features relating to the second persona of the geo-locale based on the second processed result;
computing the geo-localized persona vector based on the second plurality of features and geo-localized persona embeddings previously trained via machine learning;
generating the user persona vector by combining the hyper contextual persona vector and the geo-localized persona vector.
5 . The method of claim 4 , wherein the second plurality of features determined based on the second processing result include at least one of:
geo-specific vocabulary with words/phrases used by customers in the historic communications at the geo-locale;
geo-specific grammar used by customers in the historic communications at the geo-locale;
geo-specific emotions detected from customers in the historic communications at the geo-locale;
geo-specific demographic information associated with the geo-locale; and
geo-specific intent and strengthen thereof exhibited in the historic communications at the geo-locale.
6 . The method of claim 1 , wherein the language model is pretrained via deep learning to predict a response to a customer engaged in a communication at a geo-locale with a persona adaptively determined based on an input response input vector characterizing both a first persona of the customer and a second persona of the geo-locale.
7 . The method of claim 1 , wherein the persona-adaptive response is delivered to the customer via one of a textual form and a voice form, wherein the voice form is generated by:
creating the textual form of the persona-adaptive response; and
converting the textual form of the persona-adaptive response into a voice signal via text-to-speech transformation.
8 . A machine readable and non-transitory medium having information recorded thereon, wherein the information, when read by the machine, causes the machine to perform the following steps:
receiving a transcript of a current communication with a customer at a geo-locale;
retrieving a first set of transcripts of historic communications associated with the customer and a second set of transcripts of historic communications that occurred at the geo-locale;
computing a user persona vector based on
a hyper contextual persona vector obtained based on the transcript and the first set of transcripts, wherein the hyper contextual persona vector characterizes a first persona of the customer exhibited in the current and the historic communications, and
a geo-localized persona vector obtained based on the second set of transcripts, wherein the geo-localized persona vector characterizes a second persona of the geo-locale;
computing a current persona vector based on the transcript to characterize a current persona of the customer exhibited in the current communication;
generating, by a language model, a persona-adaptive response based on the user persona vector and the current persona vector, wherein the persona-adaptive response is expressed with a persona dynamically adapted with respect to the customer and the geo-locale; and
sending the persona-adaptive response to the customer.
9 . The medium of claim 8 , wherein the computing the user persona vector comprises:
processing the transcript and the first set of transcripts to generate a first processed result;
identifying a first plurality of features relating to the first persona of the customer based on the first processed result; and
computing the hyper contextual persona vector based on the first plurality of features and user persona embeddings previously trained via machine learning.
10 . The medium of claim 9 , wherein the first plurality of features include at least one of:
user-specific vocabulary with words/phrases used by the customer in the current and historic communications;
user-specific grammar used by the customer in the current and historic communications; and
user-specific demographic information about the customer.
11 . The medium of claim 9 , wherein the information, when read by the machine, further causes the machine to perform the following steps:
processing the second set of transcripts to generate a second processed result;
determining a second plurality of features relating to the second persona of the geo-locale based on the second processed result;
computing the geo-localized persona vector based on the second plurality of features and geo-localized persona embeddings previously trained via machine learning;
generating the user persona vector by combining the hyper contextual persona vector and the geo-localized persona vector.
12 . The medium of claim 11 , wherein the second plurality of features determined based on the second processing result include at least one of:
geo-specific vocabulary with words/phrases used by customers in the historic communications at the geo-locale;
geo-specific grammar used by customers in the historic communications at the geo-locale;
geo-specific emotions detected from customers in the historic communications at the geo-locale;
geo-specific demographic information associated with the geo-locale; and
geo-specific intent and strengthen thereof exhibited in the historic communications at the geo-locale.
13 . The medium of claim 8 , wherein the language model is pretrained via deep learning to predict a response to a customer engaged in a communication at a geo-locale with a persona adaptively determined based on an input response input vector characterizing both a first persona of the customer and a second persona of the geo-locale.
14 . The medium of claim 8 , wherein the persona-adaptive response is delivered to the customer via one of a textual form and a voice form, wherein the voice form is generated by:
creating the textual form of the persona-adaptive response; and
converting the textual form of the persona-adaptive response into a voice signal via text-to-speech transformation.
15 . A system, comprising:
a user interaction unit implemented by a processor and configured for receiving a transcript of a current communication with a customer at a geo-locale;
a persona-adaptive response generator implemented by a processor and configured for
retrieving a first set of transcripts of historic communications associated with the customer and a second set of transcripts of historic communications that occurred at the geo-locale,
computing a user persona vector based on
a hyper contextual persona vector obtained based on the transcript and the first set of transcripts, wherein the hyper contextual persona vector characterizes a first persona of the customer exhibited in the current and the historic communications, and
a geo-localized persona vector obtained based on the second set of transcripts, wherein the geo-localized persona vector characterizes a second persona of the geo-locale,
computing a current persona vector based on the transcript to characterize a current persona of the customer exhibited in the current communication, and
generating, by a language model, a persona-adaptive response based on the user persona vector and the current persona vector, wherein the persona-adaptive response is expressed with a persona dynamically adapted with respect to the customer and the geo-locale; and
the user interaction unit is further configured for sending the persona-adaptive response to the customer.
16 . The system of claim 15 , wherein the persona-adaptive response generator comprises a user persona vector generator implemented by a processor and configured for computing the user persona vector by:
processing the transcript and the first set of transcripts to generate a first processed result;
identifying a first plurality of features relating to the first persona of the customer based on the first processed result; and
computing the hyper contextual persona vector based on the first plurality of features and user persona embeddings previously trained via machine learning.
17 . The system of claim 16 , wherein the first plurality of features include at least one of:
user-specific vocabulary with words/phrases used by the customer in the current and historic communications;
user-specific grammar used by the customer in the current and historic communications; and
user-specific demographic information about the customer.
18 . The medium of claim 16 , wherein the user persona vector generator is further configured for:
processing the second set of transcripts to generate a second processed result;
determining a second plurality of features relating to the second persona of the geo-locale based on the second processed result;
computing the geo-localized persona vector based on the second plurality of features and geo-localized persona embeddings previously trained via machine learning;
generating the user persona vector by combining the hyper contextual persona vector and the geo-localized persona vector.
19 . The system of claim 18 , wherein the second plurality of features determined based on the second processing result include at least one of:
geo-specific vocabulary with words/phrases used by customers in the historic communications at the geo-locale;
geo-specific grammar used by customers in the historic communications at the geo-locale;
geo-specific emotions detected from customers in the historic communications at the geo-locale;
geo-specific demographic information associated with the geo-locale; and
geo-specific intent and strengthen thereof exhibited in the historic communications at the geo-locale.
20 . The system of claim 15 , wherein the language model is pretrained via deep learning to predict a response to a customer engaged in a communication at a geo-locale with a persona adaptively determined based on an input response input vector characterizing both a first persona of the customer and a second persona of the geo-locale.