Iterative Content Classification With Small-Label Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing classification systems for documents rely heavily on human effort, leading to inefficiencies and variations in accuracy due to operator skill, and require large amounts of training data, causing undue stress and affecting classifier accuracy.
Innovation Solution
A method utilizing machine learning with a graphical user interface that iteratively refines a learning model by repeating classification, verification, and machine learning steps, allowing operators to evaluate and add training data to improve accuracy, using a small initial dataset and incorporating unsupervised learning for clustering.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning is performed using a large amount of training data, then classifier accuracy is improved, but user burden and stress increase
Solution Approach 1:
The patent segments the training data preparation process into multiple stages: initial small dataset collection, iterative classification, user verification, and progressive data augmentation. This segmentation allows the system to achieve high accuracy without requiring users to prepare large datasets upfront, thereby reducing user burden while maintaining classifier precision.
Solution Approach 2:
The system performs preliminary machine learning with a small initial dataset to generate preliminary classification results. These results are then presented to users for verification, and only verified classifications are added to the training dataset. This preliminary action approach avoids the need for users to manually prepare extensive training data beforehand.
2Measurement precision
If machine learning is performed once with available data, then processing time is reduced, but classification accuracy is insufficient
Solution Approach 1:
The patent implements continuous machine learning where the system iteratively performs classification, receives user verification, and updates the training dataset progressively. This continuous process maintains high classification accuracy without requiring lengthy batch processing, as the model learns continuously from verified classifications rather than requiring complete retraining.
Solution Approach 2:
The system performs periodic machine learning updates at intervals determined by user verification completion rather than continuously or in large batches. This periodic action allows the system to maintain accuracy while minimizing processing time, as each learning cycle uses only the data verified up to that point rather than waiting for complete dataset preparation.
3Measurement precision
If human operators classify documents manually, then classification accuracy can be maintained, but efficiency and productivity decrease
Solution Approach 1:
The patent introduces an intermediary machine learning system that handles the bulk of classification work automatically, while human operators serve as intermediaries for verification and correction of specific classifications. This intermediary approach maintains high accuracy through human oversight while dramatically improving efficiency by automating the majority of classification tasks.
Solution Approach 2:
The system implements feedback loops where user verification of classification results feeds back into the training dataset, which then improves subsequent automated classifications. This feedback mechanism maintains high accuracy comparable to manual classification while achieving superior productivity, as the system learns from user corrections without requiring continuous manual intervention.
4Adaptability or versatility
If different operators classify the same documents, then diverse perspectives may be captured, but variation in accuracy and consistency decreases
Solution Approach 1:
The patent merges multiple operator perspectives by having multiple users verify and classify the same documents within the system. Their combined verified classifications are integrated into the training dataset, capturing diverse perspectives while maintaining consistency through the standardized verification process and centralized learning model that synthesizes all inputs.
Data Source
AI summary
A novel content classification method is provided. A content classification method using machine learning for a learning model and a classifier fabrication method are provided. In Step 1, a data set containing a plurality of contents is acquired. Learning labels are attached to m contents, and the learning labels are not attached to the remaining contents. In Step 2, a first learning model is created by machine learning using the m contents. In Step 3, judgment labels are attached to the plurality of contents using the first learning model and are displayed on a GUI. In Step 4, new learning labels are attached to k contents in the plurality of contents. In Step 5, a second learning model is created by the machine learning using the k contents. In Step 6, judgment labels are attached to the plurality of contents using the second learning model and are displayed on the GUI.