Granted Patent B2

US 12,469,010 · App. 17/080,037 · Granted Nov 11, 2025

Inventors: Srivatsan Laxman (Palo Alto, CA); Supriya A Rao (Palo Alto, CA)

G06Q10/107G06F16/22G06F16/285G06N5/025

Abstract

Claims (27)

1 . A computerized method comprising:

receiving a dialog session, wherein the dialog session comprises a set of new inbound messages;

feeding the dialog session into tokenizer;

with the tokenizer, generating a set of tokens by breaking the new inbound messages into a sequence of tokens;

providing the tokens to a DAG frame labeler cascade;

with the DAG frame labeler cascade, using a sequence of tokens to generate a set of token labels, wherein the input to the DAG frame labeler cascade is then passed to a set of levels, wherein a number of levels is dependent on a desired depth of the DAG frame, and wherein each entity group has its own level;

passing, with the DAG frame labeler cascade, the token labels and tokens to an entity interpreter;

with the entity interpreter, generating a DAG frame, wherein the entity interpreter implements an entity group alignment, wherein the entity group alignment associates a specified set of services with one or more customers from the set of new inbound messages, wherein entity interpreter implements a pronoun resolution operation;

with the DAG frame, outputting a structured information from a multiturn dialogue; and

with multitask learning framework:

receiving the DAG frame,

subject the DAG frame to a multi-task layer processing, wherein the multi-task layer processing infer a response to an incoming messages,

using the multi-task layer processing to predict various class label, wherein each prediction comes with a score that is associated with a confidence level in preparation for a set of messages that is constructed in a response, and

augmenting the DAG frame with class labels to enhance the structured annotated dialog session.

2 . The computerized method of claim 1 , wherein the structured information from multiturn dialogue represents a structure-annotated dialog session.

3 . The computerized method of claim 1 , wherein the DAG frame labeler cascade accesses a business dictionary that defines a list of staff, services, and locations.

4 . The computerized method of claim 3 , wherein the entity interpreter implements an entity to business database alignment operation where each phrase representing a detected entity in a message of the set of new inbound messages is mapped to an entry in the business menu or an inventory in relevant business database.

5 . The computerized method of claim 1 further comprising:

providing the structured information from the multiturn dialogue into a multi-task multiturn message classifier.

6 . The computerized method of claim 5 , wherein the multi-task multiturn message classifier implements workflow transition detection and frequently asked questions (FAQ) detection.

7 . The computerized method of claim 6 , wherein the workflow transition detection passes on a set of detected workflow transitions to a concatenated labeler.

8 . The computerized method of claim 7 , wherein the concatenated labeler generates a set of FAQ matches and concatenated labels.

9 . The computerized method of claim 8 , wherein the set of FAQ matches and concatenated labels are used to generate a set of predicted class labels with scores.

10 . The computerized method of claim 9 further comprising:

passing the set of predicted class labels with scores to an artificial intelligent (AI)-based business assistant.

11 . The computerized method of claim 10 , wherein the AI-based business assistant uses the set of predicted class labels with scores to update a workflow state.

12 . The computerized method for claim 10 , wherein the AI-based business assistant uses the set of predicted class labels with scores to generate a message to a customer, a business entity, or a customer support agent.

Assignments (2)

Continuity (3)

References Cited (12)