System and method for conversational middleware platform

Patent metadata

Field	Value
Publication number	US-11715465-B2
Application number	US-202117170682-A
Country	US
Kind code	B2
Filing date	Feb 8, 2021
Priority date	Feb 7, 2020
Publication date	Aug 1, 2023
Grant date	Aug 1, 2023

Abstract

A de-coupled computing infrastructure is described that is adapted to provide domain specific contextual engines based on conversational flow. The computing infrastructure further includes, in some embodiments, a mechanism for directing conversational flow in respect of a backend natural language processing engine. The computing infrastructure is adapted to control or manage conversational flows using a plurality of natural language processing agents.

First claim

What is claimed is:

1. An automated conversation orchestration system for interconnecting a plurality of natural language processing agents each having a different domain specialization or operating characteristics to generate an output response data structure responding to a new utterance string from a user, the system comprising:
one or more processors operating in conjunction with computer memory and one or more non-transitory computer readable storage mediums, the one or more processors configured to:
receive and tokenize the new utterance string from the user;
route the tokenized new utterance string to the plurality of natural language processing agents to receive one or more response confidence score values each corresponding to a corresponding natural language processing agent;
query a profile data structure associated with the user to obtain one or more probability values, each associated with a corresponding domain specialization or operating characteristic of each of the plurality of natural language processing agents;
based at least on a combination of the one or more response confidence score values and the one or more probability values, assign a primary natural language processing agent; and
generate the output response data structure using at least the assigned primary natural language processing agent;
wherein the profile data structure is a graph data structure having interconnected nodes representing bifurcated interaction decision nodes having associated weight values that have been adjusted over a set of prior recorded interactions by the user or one or more similar users; and wherein the graph data structure is traversed to obtain the one or more probability values;
wherein if two or more probability values are greater than a threshold value, an additional verification is conducted to present a bifurcated decision interaction between the two or more natural language processing agents corresponding to the two or more probability values greater than the threshold value to identify the primary natural language processing agent, and the graph data structure is updated based on the selection obtained from the additional verification such that the graph data structure is biased towards the selection in future traversals.

2. The automated conversation orchestration system of claim 1 , wherein the interaction record is discarded after updating the graph data structure, and the graph data structure is utilized across a plurality of similar users.

3. The automated conversation orchestration system of claim 1 , wherein the plurality of natural language processing agents include at least both conversational natural language processing agents and contextual natural language processing agents;
wherein the conversational natural language processing agents are configured for generating the output response data structure to have a response to be transmitted to the user or a downstream conversational natural language processing agent; and
wherein the contextual natural language processing agents are configured for triggering modifications of the new utterance string for re-processing through the automated conversation orchestration system.

4. The automated conversation orchestration system of claim 3 , wherein the contextual natural language processing agents include at least one intent tracking natural language processing agent that is configured to detect potential incongruities between an estimated intent and a literal meaning conveyed in the tokenized new utterance string, and responsive to a detection of the potential incongruities between the estimated intent and the literal meaning conveyed in the tokenized new utterance string, generate the new utterance string by replacing any tokens not aligned with the estimated intent with tokens aligned with the estimated intent such that a literal meaning of the new utterance string matches the estimated intent.

5. The automated conversation orchestration system of claim 3 , wherein the plurality of natural language processing agents further include user behavior tracking agents that are configured for tracking a set of pre-conditions, and upon determining that the pre-conditions are satisfied either in the new utterance string or in the profile data structure, insert biasing values to shift the one or more probability values.

6. The automated conversation system of claim 1 , wherein the one or more processors are configured to first replace any identified sensitive word tokens from the tokenized new utterance string with placeholder words prior to routing the tokenized new utterance string to the plurality of natural language processing agents to receive the one or more response confidence score values.

7. The automated conversation system of claim 3 , wherein the contextual natural language processing agents include at least one data sensitivity natural language processing agent that is configured for parsing individual word tokens of the tokenized new utterance string and to return a high probability if any of the individual word tokens include a word that is estimated to be sensitive such that the at least one data sensitivity natural language processing agent is selected as the primary natural language processing agent;
wherein responsive to being selected as the primary natural language processing agent, the at least one data sensitivity natural language processing agent generates a modified new utterance string to be provided to the automated conversation system for a next iteration of primary natural language processing agent selection, wherein each word that was estimated to be sensitive is replaced with a corresponding in-domain placeholder; and
wherein the output response data structure is generated replacing the corresponding in-domain placeholder with a corresponding sensitive word.

8. The automated conversation system of claim 1 , wherein the plurality of natural language processing agents are configured to interface with a set of de-coupled fulfillment handlers; and
wherein at least one natural language processing agent of the plurality of natural language processing agents, after being assigned as the primary natural language processing agent, upon estimating that the user's intent matches a capability of at least one de-coupled fulfillment handler of the set of de-coupled fulfillment handlers at a high level of confidence, invokes the corresponding de-coupled fulfillment handler to initiate a new data process representative of an automated task by passing in at least one parameter extracted from or based on individual word tokens of the tokenized utterance string.

9. The automated conversation system of claim 3 , wherein the primary conversational natural language processing agent, upon detecting an intent in a new utterance string to change a domain specialization such that another conversational natural language processing agent will be assigned as the primary conversational natural language processing agent, generates an agent hand-off utterance string for routing by a next iteration of routing by the automated conversation system.

10. The automated conversation system of claim 3 , wherein the primary conversational natural language processing agent, upon detecting an intent in a new utterance string to include another user generates one or more user hand-off utterance strings for routing by another instance of the automated conversation system;
wherein the one or more user hand-off utterance strings includes one or more additional contextual data objects applicable to the another instance of the automated conversation system.

11. The automated conversation system of claim 1 , wherein the output response data structure is lin

Assignees

• Royal Bank Of Canada

Inventors

• Ahmadidaneshashtiani Mohammadhosein
• Middleton Ian Robert
• Munro Shawn Harold
• Macnamara Darren Michael
• Sang Bo
• Jaiswal Devina
• Liu Hanke
• To Kylie

Classifications

• G10L15/19 Primary

  Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules · CPC title

• G06F40/284

  Lexical analysis, e.g. tokenisation or collocates · CPC title

• H04L51/02 Primary

  using automatic reactions or user delegation, e.g. automatic replies or chatbot-generated messages · CPC title

• H04L51/066

  Format adaptation, e.g. format conversion or compression · CPC title

• H04L51/222

  using geographical location information, e.g. messages transmitted or received in proximity of a certain spot or area · CPC title