Dialogue State Tracking With Summary-Based Model Training

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing dialogue state tracking technologies using statistical language models require large amounts of training data and significant computational resources due to the need for generating and verifying numerous question-answer pairs, limiting their effectiveness.

Innovation Solution

A method and system for tracking dialogue states using an artificially generated dialogue summary sentence as a training set, employing a dialogue state tracking model with an input layer, output layer, and hidden layer to generate a dialogue state template.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If statistical language models are used for dialogue state tracking, then the model can interpret dialogue state information, but a considerable amount of training data and computational resources are required

Engineering Contradiction:
Improvedialogue state tracking accuracyVSAvoidtraining data volume
Core Design Contradiction:
ReliabilityVSQuantity of substance

Solution Approach 1:

The patent applies preliminary action by pre-generating dialogue summary sentences that capture essential dialogue state information before training the model. These pre-generated summaries serve as prepared training materials that eliminate the need for extensive manual annotation and large-scale data collection, directly resolving the contradiction between tracking accuracy and data volume requirements

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces dialogue summary sentences as an intermediary element between raw dialogue data and model training. These summaries act as a mediator that condenses complex dialogue information into structured, meaningful representations, enabling effective model training with significantly reduced data requirements while maintaining tracking accuracy

Inventive Principle:
Principle #24Intermediary (Mediator)

2Productivity

If question-answer systems are used to track dialogue state, then information can be directly queried, but a considerable amount of computation is required for generating and verifying question-answer pairs

Engineering Contradiction:
Improveinformation retrieval efficiencyVSAvoidcomputational power
Core Design Contradiction:
ProductivityVSPower

Solution Approach 1:

The patent extracts essential dialogue state information directly from dialogue data through automated summary generation, eliminating the need to create and process numerous question-answer pairs. This extraction approach retrieves necessary information efficiently without the computational overhead of generating and verifying extensive Q&A datasets

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent creates condensed copies of dialogue information in the form of dialogue summary sentences that capture key state information. These summaries serve as efficient representations that can be processed with minimal computational resources, replacing the need for complex question-answer generation and verification processes

Inventive Principle:
Principle #26Copying

Data Source

PatentUS12406148B2Method, device, and system for tracking dialogue state
Publication Date: 2025.09.02 RIIID CO
  • US12406148B2 patent drawing
  • US12406148B2 patent drawing
  • US12406148B2 patent drawing

AI summary

A method of tracking a dialogue state according to an embodiment of the present application includes: acquiring a trained dialogue state tracking model; acquiring target dialogue data; acquiring dialogue summary data from the target dialogue data using the dialogue state tracking model; and generating a dialogue state template from the dialogue summary data, in which the dialogue state tracking model includes an input layer for receiving the target dialogue data, an output layer for outputting the dialogue summary data, and a hidden layer having a plurality of nodes connecting the input layer and the output layer, and is trained using a training set that includes dialogue data and a dialogue summary sentence generated from dialogue state data related to the dialogue data.