Dynamic Topic Transfer Graphs for Coherent Dialogue Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing human-computer dialogue systems struggle to actively guide topics and maintain coherence in multi-round conversations, leading to inflexible and incoherent interactions.
Innovation Solution
A conversation content generation method using a preset topic transfer graph with nodes and connecting lines representing transfer probabilities, determining a reply topic based on the current utterance, target topic, and generating content accordingly, while considering knowledge graphs for informative responses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a task-oriented dialogue system with modular structure is used, then the system is easy to implement and has strong interpretability, but the system is inflexible and difficult to adapt to varied application scenarios
Solution Approach 1:
The patent applies dynamics by transforming the static modular structure into a dynamic topic transfer graph where nodes and edges can be flexibly added, removed, and adjusted based on different application scenarios. The graph structure allows the system to adapt its dialogue flow dynamically while maintaining the ease of implementation through a visualizable framework.
Solution Approach 2:
The patent uses parameter changes by modifying the topic transfer graph through adding or removing nodes and edges, adjusting transfer probabilities, and changing graph structure to adapt to different application scenarios. This allows the same base system to serve multiple purposes by simply changing graph parameters rather than rewriting code.
2Ease of operation
If existing generative models or seq2seq architectures are used for chit-chat dialogue, then the system can generate natural responses, but the coherence of the dialogue needs to be improved
Solution Approach 1:
The patent introduces the topic transfer graph as an intermediary between the user input and the response generation. The graph acts as a mediator that guides the dialogue flow by determining which topics should be transferred and with what probability, thereby improving coherence while preserving the naturalness of generative model responses.
Solution Approach 2:
The patent implements feedback by using the topic transfer graph to continuously monitor and guide the dialogue state. The graph provides feedback signals about which topics are currently active and which should be transferred, creating a closed-loop system that maintains coherence across multiple dialogue turns.
3Device complexity
If the computer responds passively to user questions, then the implementation is simple, but the computer cannot actively and naturally guide topics in conversation
Solution Approach 1:
The patent applies preliminary action by pre-defining the topic transfer graph structure with nodes representing topics and edges representing possible transfers. This preparation is done beforehand, allowing the system to actively guide conversations through pre-planned topic pathways without requiring complex real-time decision-making algorithms.
Solution Approach 2:
The topic transfer graph serves as an intermediary that enables the computer to actively guide topics. Instead of directly controlling the conversation flow through complex logic, the graph mediates by providing a structured framework that naturally guides the dialogue toward desired topics through probability-based transitions.
Data Source
AI summary
A conversation content generation method and apparatus, a storage medium and a terminal are provided. The method includes: acquiring a current utterance entered by a user; reading a preset topic transfer graph and target topic, wherein the topic transfer graph includes nodes and connecting lines between the nodes, the nodes correspond to topics in one-to-one correspondence, each connecting line points from a first node to a second node, a weight of the connecting line indicates probability of transferring from a topic corresponding to the first node to a topic corresponding to the second node, and the topic transfer graph includes a node corresponding to the target topic; determining a topic of reply content of the current utterance at least based on the current utterance, the topic transfer graph and the target topic, and recording it as a reply topic; generating the reply content at least based on the reply topic.