Text Q&A Model Training Using Uncertainty-Based Question Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing text question and answer (Q&A) models face challenges in training efficiency due to varying difficulty levels of question texts, leading to poor performance in learning from both simple and complex questions.
Innovation Solution
A method for training a text Q&A model that involves determining sample question texts and answers, calculating uncertainty degrees based on prediction probabilities, and adjusting model parameters to focus on learning difficult questions, accompanied by filtration of simpler questions to enhance training efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the Q&A model is trained using conventional methods on all question texts, then the model can learn from both simple and complex questions, but the training efficiency is poor due to varying difficulty levels of questions
Solution Approach 1:
The patent changes the parameter of training sample selection by filtering out simple questions based on uncertainty degree thresholds, transforming the training process from uniform treatment of all questions to selective processing based on difficulty parameters
Solution Approach 2:
The patent applies different processing strategies to different parts of the question set: simple questions (below threshold) are filtered out, while complex questions (above threshold) are retained for training, creating local quality differentiation in the training data
2Productivity
If the model focuses on complex questions by filtering simple questions, then the training efficiency improves, but the quantity of training data decreases
Solution Approach 1:
The patent extracts and removes simple questions from the training set based on uncertainty degree calculation, keeping only the complex questions that provide meaningful learning signals for the model
Solution Approach 2:
The patent generates multiple versions of complex questions through text augmentation techniques, creating synthetic training samples that increase data volume while maintaining quality
Data Source
AI summary
A method for training a text question and answer (Q&A) model is performed by an electronic device. The method includes: determining a sample question text set and a sample answer text corresponding to a sample question text in the sample question text set; inputting the sample question text into a text Q&A model to be trained, and obtaining a predicted answer text output by the text Q&A model and at least one prediction probability of at least one reference character on each character position in the predicted answer text; determining an uncertainty degree of the predicted answer text; and obtaining a trained text Q&A model by adjusting a parameter of the text Q&A model based on the sample answer text, the predicted answer text and the uncertainty degree of the predicted answer text.