Multi-Domain Knowledge Graph Inference for Tactful Dialog
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in generating dialog that is perceived as 'tactful' by humans, particularly in dynamic environments where information changes frequently, making it difficult to manually design dialog sequences or collect comprehensive dialog data for machine learning.
Innovation Solution
An inference device and method that integrate information from multiple domains using graph-based techniques, combining directed graphs to generate an integrated graph, and calculating importance levels of nodes using random walk probabilities or PageRank algorithms to derive suitable inference results for users.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If manual design of dialog sequences is used, then dialog can be customized for specific situations, but it becomes difficult to cover all dynamic situations and requires extensive human effort
Solution Approach 1:
The system automatically generates dialog sequences by performing inference on integrated information from multiple domains, eliminating the need for manual design of dialog sequences for every possible situation. The inference device autonomously determines appropriate dialog responses based on current context.
Solution Approach 2:
The dialog system dynamically adapts to changing situations by continuously integrating information from multiple domains and performing real-time inference, allowing the dialog sequence to change flexibly based on current context rather than following fixed manual designs.
2Extent of automation
If end-to-end supervised learning is performed on dialog data, then the system can learn from data, but collecting comprehensive dialog data covering all cases requires extremely high cost or is practically impossible
Solution Approach 1:
The inference device acts as an intermediary that generates dialog sequences without requiring large amounts of annotated dialog data. It integrates information from multiple domains and performs logical inference to generate appropriate responses, bypassing the need for extensive supervised learning datasets.
Solution Approach 2:
The system replaces the traditional machine learning approach that requires large datasets with an inference-based approach using knowledge graphs and logical reasoning, eliminating the dependency on extensive dialog data collection.
3Adaptability or versatility
If information from multiple domains is integrated, then tactful dialog can be achieved, but the complexity of integrating and processing information from multiple domains increases
Solution Approach 1:
The system merges information from multiple domains into a unified knowledge graph structure, where information from different sources is integrated through standardized nodes and edges, simplifying the processing complexity while maintaining comprehensive information integration.
Solution Approach 2:
The inference device uses a universal inference mechanism that can process information from any domain through the same knowledge graph integration approach, allowing multi-domain information integration without proportionally increasing processing complexity.
Data Source
AI summary
An inference device includes processing circuitry to generate integrated information by combining items of information respectively belonging to domains different from each other included in dynamically changing external information by means of forward chaining by using knowledge information including information regarding a human's condition and information regarding a human's action and provided from a knowledge base and an inference rule provided from a rule database. Each of the items of information respectively belonging to domains different from each other is information that can be represented as a directed graph including a node and an edge. The integrated information is an integrated graph. The processing circuitry calculates an importance level of a node as a component of the integrated graph from a probability of arriving at the node in a stationary state reached by performing a random walk on the integrated graph or by using an algorithm of PageRank.