Text Classification Training Set Optimization via Iterative Error Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing text classification algorithms are affected by incorrectly tagged samples and insufficient data, leading to poor performance, as conventional methods either increase manual tagging costs or fail to correct noisy data effectively.
Innovation Solution
A method and device for optimizing the training set by selecting and correcting incorrectly tagged samples using an active learning approach, involving random selection, manual correction, and data enhancement techniques such as n-gram models and TF-IDF algorithms to improve data quality and reduce manual tagging costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manually correct all incorrectly tagged samples in the training set, then data quality and classification performance are improved, but manual tagging cost and time consumption increase significantly
Solution Approach 1:
The system uses the trained text classification model to automatically predict and identify incorrectly tagged samples by comparing model predictions with existing tags. The model serves itself to detect errors in the training data, reducing the need for manual inspection of all samples while maintaining high data quality standards.
Solution Approach 2:
The system implements an iterative feedback loop where the text classification model is trained on the training set, then used to predict samples and generate a prediction result. Incorrectly tagged samples are identified by comparing predictions with actual tags, and the training set is updated with corrected samples. This feedback mechanism continuously improves data quality without requiring manual review of every sample.
2Reliability
If increase the number of tagged samples in the training set, then classification performance is improved, but manual tagging cost increases
Solution Approach 1:
The system leverages the trained text classification model to automatically tag new samples by generating prediction results. The model applies learned patterns from the training set to classify new data, eliminating the need for manual tagging of additional samples while maintaining consistent classification performance and scaling efficiently.
3Ease of manufacture
If use conventional text classification algorithms with noisy training data, then implementation is simple, but classification performance deteriorates
Solution Approach 1:
The system performs preliminary identification and correction of incorrectly tagged samples before final model training. By using the trained model to predict samples and identify errors in advance, the system prepares a cleaned training set that eliminates noisy data, thereby improving classification performance while maintaining implementation simplicity through automated processes.
Data Source
AI summary
A method for optimizing a training set for text classification includes: the training set for text classification is acquired; part of samples are selected from the training set as a first initial training subset, and an incorrectly tagged sample in the first initial training subset is corrected to obtain a second initial training subset; a text classification model is trained according to the second initial training subset; the samples in the training set are predicted by the trained text classification model to obtain a prediction result; an incorrectly tagged sample set is generated according to the prediction result; a key incorrectly tagged sample is selected from the incorrectly tagged sample set, and a tag of the key incorrectly tagged sample is corrected to generate a correctly tagged sample corresponding to the key incorrectly tagged sample; and the training set is updated by using the correctly tagged sample.