LSTM Bot State Inference via Masking Actions

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for creating and maintaining goal-oriented dialogue systems, such as transactional bots, face challenges in ease of use, maintenance, and automatic improvement, particularly due to the complexity of rule-based systems and the need for manual intervention in supervised and reinforcement learning processes.

Innovation Solution

The system employs a Long Short-Term Memory (LSTM) recurrent neural network optimized through supervised and reinforcement learning, allowing for autonomous improvement and flexible configuration without requiring predefined state variables, and enabling developers to specify bot behavior using example dialogs and masking actions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If rule-based systems are used for dialogue systems, then the system structure is clear and controllable, but the complexity of creation and maintenance increases significantly

Engineering Contradiction:
Improvesystem controllabilityVSAvoidsystem creation and maintenance complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent replaces rule-based mechanical systems with a neural network-based learning system. The dialogue system uses neural networks to automatically learn conversation patterns and rules from data, eliminating the need for manual rule creation and maintenance while maintaining system controllability through the learned models.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The dialogue system performs self-learning and self-improvement through automated training on conversation data. The system automatically updates its knowledge base and refines its responses without requiring manual intervention for rule updates, reducing maintenance complexity while maintaining reliable performance.

Inventive Principle:
Principle #25Self-service

2Extent of automation

If supervised learning and reinforcement learning are implemented, then the bot can learn and improve autonomously, but the need for manual intervention and expert knowledge increases

Engineering Contradiction:
Improveautonomous learning capabilityVSAvoidmanual intervention requirement
Core Design Contradiction:
Extent of automationVSEase of operation

Solution Approach 1:

The patent introduces an automated training pipeline that serves as an intermediary between raw conversation data and the neural network model. This pipeline automatically handles data preprocessing, model training, and evaluation, reducing the need for expert manual intervention while enabling autonomous learning through structured automated processes.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If predefined state variables are required, then the dialog state tracking is precise, but the flexibility and adaptability of the system decreases

Engineering Contradiction:
Improvedialog state tracking precisionVSAvoidsystem flexibility
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent implements dynamic state representation where the dialogue state is inferred automatically by the neural network based on conversation context rather than relying on fixed predefined variables. This allows the system to adapt to different conversation scenarios dynamically while maintaining precise state tracking through the learned representations.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3455795B1Deep learning of bots through examples and experience
Publication Date: 2025.01.01 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP3455795B1 patent drawingFigure 1~2
  • EP3455795B1 patent drawingFigure 3
  • EP3455795B1 patent drawingFigure 4

AI summary

Embodiments include methods and systems for using, creating and maintaining goal-oriented, dialog systems (i.e., transactional bots) that provide interfaces to application functionality. The methods and systems of the embodiments provide a bot that may learn in supervised learning and reinforcement learning from conversational examples provided by domain experts and from interaction with users. Conversational bots may be created to interact using both text and/or application programming interface (API) calls. A developer may configure a bot that interfaces with an application back-end where behavior of the bot may be controlled by use of masking actions. A specification for the bot may be flexibly designed to specify how developer code may be organized, for example, as masking operations on the possible actions the bot may execute. Additionally, the methods and systems may automatically infer the best state representation during a dialog so a state variable need not be predefined.