Dialogue System Ontology Mapping for Multi-Domain Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current dialogue systems require significant training, maintenance, and human design input to operate across multiple domains, which is inefficient and labor-intensive.
Innovation Solution
A dialogue system architecture that uses a state tracker model, identifier model, and policy model to dynamically update and adapt system states, identify relevant categories, and generate actions across multiple pre-defined domain-specific ontologies, allowing for ontology-independent operation and efficient sub-domain creation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If dialogue systems use pre-defined domain-specific ontologies and policy models for each domain, then they can operate in specific domains with defined functionality, but they require significant training, maintenance, and human design input to work across multiple domains
Solution Approach 1:
The patent applies universality by using a single policy model that can handle multiple domains through ontology mapping. Instead of maintaining separate policy models for each domain, the system uses one universal policy model that receives ontology-mapped inputs, allowing it to function across different domains without requiring domain-specific training or maintenance for each domain.
Solution Approach 2:
The patent introduces an intermediary ontology mapping layer between the input data and the policy model. This mapping layer transforms domain-specific ontologies into a unified representation that the single policy model can process, acting as a mediator that enables multi-domain operation without requiring the policy model to be retrained for each domain.
2Ease of operation
If dialogue systems switch between pre-defined domain ontologies using topic trackers, then they can identify the most closely matching domain for input, but they still require extensive human design input and training for each domain
Solution Approach 1:
The patent applies self-service by enabling the system to automatically adapt to new domains through the ontology mapping mechanism. When a new domain is introduced, the system automatically maps its ontology to the unified representation without requiring human designers to manually configure or train the policy model for that domain, reducing human design input while maintaining automatic domain identification.
Data Source
AI summary
A dialogue system including: an input receiving data relating to a speech or text signal originating from a user; and a processor configured to: update a system state based on the input data using a state tracker model, the system state including probability values associated with each of plural possible values for each of plural categories; identify one or more relevant categories based on at least part of the updated system state information using an identifier model; define a set of information from stored information including plural action functions and categories, excluding categories not identified as relevant; generate a reduced system state, including the probability values associated with one or more of the plural possible values for each relevant category; determine an action based on the reduced system state and the set of information using a policy model; output information specified by the determined action at an output.