Dialogue Generation System with Preliminary Knowledge Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current human-machine dialogue systems struggle to generate high-quality dialogue responses due to the lack of necessary posterior information, leading to generic, short, and less informative responses.
Innovation Solution
The proposed solution involves a dialogue generation method that predicts a preliminary dialogue response based on a dialogue preceding text and candidate knowledge texts. This method processes the dialogue preceding text to obtain a vector representation, selects a target knowledge text based on probability values, and generates a final dialogue response incorporating the selected knowledge.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a dialogue response is generated without using external knowledge, then the generation process is simple and fast, but the response quality is poor and lacks information
Solution Approach 1:
The system performs preliminary knowledge selection before dialogue generation by predicting which knowledge items are most relevant to the user's query. This preliminary action filters and prepares knowledge in advance, allowing the generation process to focus on creating responses rather than searching for relevant information during the generation step.
Solution Approach 2:
The patent introduces an intermediary knowledge selection mechanism that acts as a mediator between the user query and the dialogue generation process. This intermediary component (knowledge selection network) processes the query, selects relevant knowledge items, and passes them to the generation model, thereby improving response quality without directly complicating the core generation function.
2Loss of information
If multiple knowledge items are selected for dialogue generation, then the response becomes more informative, but the probability distribution calculation becomes complex
Solution Approach 1:
Instead of considering all possible knowledge items or using complex methods to evaluate every combination, the system uses partial action by selecting only the top-k most relevant knowledge items based on predicted probability values. This approach provides sufficient information for high-quality responses while significantly simplifying the computation compared to exhaustive methods.
Solution Approach 2:
The system changes the parameter of knowledge item selection from uniform or random selection to probability-based selection. By using predicted probability values to weight and rank knowledge items, the system achieves more informative responses while managing computational complexity through parameter transformation rather than increasing computational depth.
3Measurement precision
If the dialogue system uses posterior information for knowledge selection, then the accuracy of knowledge selection improves, but the system requires more computational resources
Solution Approach 1:
The system performs preliminary computation of probability values for knowledge item selection before the actual dialogue generation. By pre-calculating and storing these probability distributions, the system can quickly access and use posterior information during generation without requiring extensive real-time computational resources, thus improving accuracy while managing energy consumption.
Data Source
AI summary
A dialogue generation method, a network training method and apparatus, a storage medium, and a device are provided. The method includes: predicting, based on a plurality of a plurality of pieces of candidate knowledge text in a first candidate knowledge set, a preliminary dialogue response of a first dialogue preceding text; processing the first dialogue preceding text based on the preliminary dialogue response to obtain a first dialogue preceding text vector; obtaining a piece of target knowledge text based on a probability value of the piece of target knowledge text of being selected to be used in generating a final dialogue response, the probability value being obtained based on the first dialogue preceding text vector; and generating the final dialogue response based on the first dialogue preceding text and the piece of target knowledge text.