US 10,891,438 B2
Deep learning techniques based multi-purpose conversational agents for processing natural language queries
Mahesh Prasad Singh, Noida (IN); Puneet Agarwal, Noida (IN); Ashish Chaudhary, Noida (IN); Gautam Shroff, Noida (IN); Prerna Khurana, Noida (IN); Mayur Patidar, Noida (IN); Vivek Bisht, Noida (IN); Rachit Bansal, Noida (IN); Prateek Sachan, Noida (IN); and Rohit Kumar, Noida (IN)
Assigned to Tata Consultancy Services Limited, Mumbai (IN)
Filed by Tata Consultancy Services Limited, Mumbai (IN)
Filed on Apr. 15, 2019, as Appl. No. 16/384,316.
Claims priority of application No. 201821014473 (IN), filed on Apr. 16, 2018.
Prior Publication US 2019/0317994 A1, Oct. 17, 2019
Int. Cl. G06F 40/30 (2020.01); G06N 20/00 (2019.01)
CPC G06F 40/30 (2020.01) [G06N 20/00 (2019.01)] 15 Claims
OG exemplary drawing
 
1. A method of Deep Learning techniques based multi-purpose conversational agents for processing natural language queries, the method being implemented by one or more processors and comprising:
defining, by the one or more processors, a plurality of components comprising a Dialogue State Manager (DSM), a Multi-level Intent Identification Component, an Agents Manager, a plurality of Primary Agents, an Intent-Action-Dialogue (IAD) Framework, a Query-Update-Engage (QUE) Framework, a Knowledge Graph Update-Natural Language (KGU-NL) Agent, a Knowledge Graph Engage Agent, a plurality of Auxiliary Agents, and a Knowledge Graph Update Agent, wherein each component amongst the plurality of components comprises one or more multi-purpose conversational agents;
logically integrating, based upon a set of anticipated natural language user queries, the plurality of components by one or more application programming interfaces (APIs);
receiving, by the plurality of components logically integrated, a set of natural language queries from a plurality of sources;
performing, based upon the set of natural language queries, a plurality of steps, wherein the plurality of steps comprise:
(i) identifying at least one multi-purpose conversational agent amongst the one or more multi-purpose conversational agents by using the DSM, wherein the identified multi-purpose conversational agent corresponds to either the IAD Framework or the QUE Framework; and
(ii) predicting, by using one or more Deep Learning techniques, a probable user intent against a user query amongst the set of non-classified natural language queries;
performing, based upon the predicted user intent and the identified multi-purpose conversational agent, steps of:
(i) selecting one or more pre-defined set of responses amongst a plurality of pre-defined set of responses or engaging a user for extracting in-depth information or calling external APIs for communicating the in-depth information to one or more external services upon determining the identified multi-purpose conversational agent to be corresponding to the IAD Framework, wherein the one or more pre-defined set of responses and the in-depth information correspond to the set of natural language queries;
(ii) classifying a query amongst the set of natural language queries to identify one or more categories of conversations by implementing a recurrent neural network technique upon determining the identified multi-purpose conversational agent to be corresponding to the QUE Framework; and
(iii) performing, based upon the classified query, steps of:
(a) querying one or more knowledge graphs to generate a first set of responses corresponding to the set of natural language queries; and
(b) updating, by the KGU-NL Agent, the one or more knowledge graphs to generate a second set of responses corresponding to the set of natural language queries,
wherein the one or more knowledge graphs are updated by the Knowledge Graph Update Agent based upon a set of information obtained from one or more users by the Knowledge Graph Engage Agent, and wherein the set of information corresponds to the processing of natural language queries.