Dynamic Memory Network for Intent Recognition in Chatbots
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current AI chatbots face challenges in generating effective responses due to difficulties in deriving attention points and intentions from user messages, especially when emotions and facts are expressed implicitly or in a dispersed manner.
Innovation Solution
The proposed solution employs a dynamic memory network (DMN) to determine intention vectors based on fact and emotion memory vectors, using sentiment analysis classifiers and knowledge graphs to generate responses that extend attention points in accordance with user intentions, and utilizes a generative adversarial network (GAN) for refining responses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If keyword scanning or simple natural language processing is used to generate responses, then the response generation process is fast and simple, but the accuracy and contextual relevance of responses deteriorate when emotions and intentions are implicit or dispersed
Solution Approach 1:
The patent segments the intention recognition process into multiple independent components: emotion detection module, fact extraction module, attention point identification module, and intention inference module. Each module processes specific aspects of user input separately, then integrates results to improve overall accuracy without significantly increasing processing time
Solution Approach 2:
The patent introduces an intermediary dynamic memory network that acts as a bridge between raw user input and final response generation. This intermediary layer processes and structures implicit emotions and intentions before they reach the response generation module, improving accuracy while maintaining reasonable processing speed
2Measurement precision
If dynamic memory network with sentiment analysis and knowledge graphs is used to capture implicit emotions and intentions, then the accuracy and contextual relevance of responses is improved, but the system complexity increases
Solution Approach 1:
The patent designs a universal dynamic memory network structure that performs multiple functions: sentiment analysis, fact extraction, attention tracking, and intention inference. This multi-functional architecture reduces overall system complexity by consolidating what could be separate specialized modules into one integrated framework
Solution Approach 2:
The patent implements preliminary action by pre-processing user input through emotion detection and fact extraction before main intention recognition. Knowledge graphs are pre-built and prepared in advance, allowing the system to quickly query and match patterns during conversation without complex real-time processing
Data Source
AI summary
The present disclosure provides method and apparatus for generating responses in automated chatting. A message may be received in a session. An intention vector may be determined based at least on the message and the session through dynamic memory network (DMN), the intention vector indicating an attention point and an intention. A response may be generated based at least on the intention vector.