Conversation Processing System Using Configurable State Machine

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current task-oriented conversation systems require significant redevelopment efforts and lack flexibility when adapting to new tasks, with existing methods either requiring extensive customization or relying on complex reinforcement learning that is difficult to control and understand.

Innovation Solution

A conversation processing method and system that acquires task-related user requirements, updates user conversation states using pre-configured task parameters with multiple slots, and generates candidate actions and response text, allowing for customization of conversation services without full redevelopment, thereby improving expandability and reducing development costs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If each module is redeveloped according to the service logic of the specific task, then the system can meet the requirement of the specific task well, but the development effort is large

Engineering Contradiction:
Improvetask requirement fulfillmentVSAvoiddevelopment effort
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent creates a universal conversation management system with standardized modules (state expression module, system action module, dialogue flow control module) that can handle multiple different tasks through configuration rather than redevelopment. The state expression module uses a unified state machine framework that adapts to different tasks by configuring transition conditions and actions, enabling the same system architecture to serve flight booking, hotel reservation, and other domains without requiring complete module redevelopment for each task.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The conversation management system is divided into independent, modular components: the state expression module, system action module, and dialogue flow control module. Each module has a specific function and can be configured independently. The state expression module contains state definitions, transition conditions, and actions as separate configurable elements, allowing the system to be assembled and adapted for different tasks by configuring these segments rather than rewriting entire modules.

Inventive Principle:
Principle #1Segmentation

2Adaptability or versatility

If a learning reinforcement method is used to construct a conversation system via environmental interaction learning, then the system can be constructed with abstracted state expression and system action, but a large amount of conversation materials is needed and the understandability of the learnt strategy is poor

Engineering Contradiction:
Improveconversation system constructionVSAvoidimplementation difficulty
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent introduces a standardized state expression module as an intermediary between the task logic and the conversation system. This module provides a formal, structured way to represent conversation states, transition conditions, and actions using a state machine framework. Instead of relying on black-box reinforcement learning, the intermediary module allows developers to explicitly define and control conversation flow through configurable parameters, improving both understandability and implementability while maintaining adaptability to different tasks.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Adaptability or versatility

If a learning reinforcement method is used to construct a conversation system, then the system can be constructed with abstracted state expression and system action, but the controllability of the system is low

Engineering Contradiction:
Improveconversation system constructionVSAvoidsystem controllability
Core Design Contradiction:
Adaptability or versatilityVSEase of operation

Solution Approach 1:

The patent implements a dynamic state machine framework where the conversation system can adapt its behavior based on configured transition conditions and actions. The state expression module allows dynamic switching between different conversation states based on user input and system context, with controllable transition rules. This provides both adaptability to different tasks and controllability through explicit configuration of state transitions, avoiding the uncontrollable nature of pure reinforcement learning approaches.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3096226B1Conversation processing method and computer device
Publication Date: 2020.03.04 BAIDU ONLINE NETWORK TECH (BEIJIBG) CO LTD
  • EP3096226B1 patent drawingFigure 1
  • EP3096226B1 patent drawingFigure 2~3
  • EP3096226B1 patent drawingFigure 4~5

AI summary

The present invention provides a conversation processing method, a conversation management system and a computer device. The method comprises: acquiring task-related user requirement data; updating a user conversation state based on the user requirement data and a pre-configured task-related task parameter; generating at least one piece of candidate action data according to the user conversation state; and generating response text data according to the at least one piece of candidate action data. The expandability of the conversation management system is improved by customizing the conversion service related to the task in the conversation system.