Dialogue belief tracking via probabilistic state graph inference

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current spoken dialogue systems are limited in their ability to infer user goals or intents over large multi-domain information repositories, especially when these goals or intents are not explicitly stated, and they struggle with accuracy and versatility in handling complex user requests.

Innovation Solution

The system utilizes a dialogue belief tracking method that maps the dialogue state onto a large graphical knowledge base, creating a probabilistic model to infer user goals and intents, allowing for multiple intent inference and ambiguity resolution across multiple conversation turns without requiring manual design of graphical models.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual design of graphical models is used for dialogue state tracking, then model accuracy can be optimized, but system complexity and development time increase significantly

Engineering Contradiction:
Improvedialogue state tracking accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system automatically learns dialogue state tracking models from conversation data without requiring manual graphical model design. The machine learning algorithm self-adjusts parameters and structures based on training data, enabling the system to serve itself in creating accurate dialogue representations while reducing human intervention and complexity.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent transforms the fixed manual graphical model structure into dynamic parameters that can be automatically adjusted through machine learning. By changing from static manually-designed graphs to learned parameter-based models, the system achieves adaptability while reducing the burden of manual model construction and maintenance.

Inventive Principle:
Principle #35Parameter changes

2Ease of operation

If the system infers user goals from implicit statements, then user interaction becomes more natural, but accuracy in determining user intent decreases

Engineering Contradiction:
Improveuser interaction naturalnessVSAvoiduser intent determination accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The system incorporates feedback mechanisms where dialogue state tracking results are continuously refined based on user responses and conversation context. The machine learning model learns from feedback signals to improve its inference accuracy over time, allowing the system to handle implicit user goals more reliably while maintaining natural interaction.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The system performs preliminary dialogue state tracking and intent classification before final action execution. By pre-processing and analyzing conversation context in advance, the system prepares multiple potential user intents and ranks them, improving accuracy in determining the correct user goal even when statements are implicit.

Inventive Principle:
Principle #10Preliminary action

3Adaptability or versatility

If the system handles multiple domains and complex user requests, then versatility improves, but reliability in accurate inference deteriorates

Engineering Contradiction:
Improvemulti-domain handling capabilityVSAvoidinference reliability
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent segments the dialogue understanding task into multiple specialized components including entity recognition, intent classification, and dialogue state tracking. Each component focuses on specific aspects of user input, allowing the system to handle multiple domains effectively while maintaining reliability through specialized processing for each segment.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The machine learning-based dialogue state tracker serves as a universal component that can handle multiple domains and task types simultaneously. By creating a multi-functional model that learns from diverse conversation data, the system achieves versatility across domains while maintaining consistent inference reliability through unified processing architecture.

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

Data Source

PatentEP3230979B1Method and system for determining user intent in a spoken dialog based on transforming at least one portion of a semantic knowledge graph to a probabilistic state graph
Publication Date: 2018.09.26 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP3230979B1 patent drawingFigure 1
  • EP3230979B1 patent drawingFigure 2
  • EP3230979B1 patent drawingFigure 3A~3D

AI summary

Systems and methods for responding to spoken language input or multi-modal input are described herein. More specifically, one or more user intents are determined or inferred from the spoken language input or multi-modal input to determine one or more user goals via a dialogue belief tracking system. The systems and methods disclosed herein utilize the dialogue belief tracking system to perform actions based on the determined one or more user goals and allow a device to engage in human like conversation with a user over multiple turns of a conversation. Preventing the user from having to explicitly state each intent and desired goal while still receiving the desired goal from the device, improves a user's ability to accomplish tasks, perform commands, and get desired products and/or services. Additionally, the improved response to spoken language inputs from a user improves user interactions with the device.