Target Domain Classifier Training via Iterative Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning-based systems for cross-domain text classification assume consistent data distributions and sentiments between source and target domains, which often fail due to inconsistencies in real-world data.
Innovation Solution
A method and system for training a target domain classifier that identifies common keywords with consistent labels across source and target domains, iteratively trains classifiers to label text segments, and re-trains based on high-confidence segments to adapt to domain-specific features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional machine learning systems assume consistent data distributions between source and target domains, then training is simplified, but classification accuracy deteriorates due to real-world data inconsistencies
Solution Approach 1:
The patent segments the training process into multiple iterative stages: initial training on source domain data, then progressive refinement using identified target domain samples. This segmentation allows the system to first establish a baseline model and then progressively adapt to domain-specific characteristics, resolving the contradiction between training simplicity and accuracy.
Solution Approach 2:
The system implements feedback mechanisms where classification results are continuously evaluated, and high-confidence target domain samples are fed back into the training process. This feedback loop enables the model to progressively improve accuracy by learning from its own predictions and adjusting to domain shifts, while maintaining manageable training complexity through automated feedback-driven iteration.
2Measurement precision
If manual labeling of target domain data is performed to achieve high accuracy, then classification precision improves, but time consumption and costs increase
Solution Approach 1:
The system enables self-service by automatically identifying high-confidence target domain samples and using them for iterative re-training without manual intervention. The automated confidence-based selection and re-training process allows the system to improve its own performance autonomously, achieving high classification precision while eliminating time-consuming manual labeling efforts.
Solution Approach 2:
The system performs preliminary action by pre-identifying high-confidence samples from the target domain before full-scale labeling would be required. By proactively selecting and utilizing these confident samples for iterative training, the system achieves accurate classification without the need for comprehensive manual labeling of all target domain data.
3Adaptability or versatility
If iterative re-training with high-confidence samples is performed, then model adaptability to target domain improves, but computational resources increase
Solution Approach 1:
The system applies partial action by selectively re-training only with high-confidence target domain samples rather than using all available data in each iteration. This selective approach achieves effective domain adaptability by focusing computational resources on the most informative samples, thereby reducing overall computational energy consumption while maintaining strong adaptability to the target domain.
Data Source
AI summary
The disclosed embodiments illustrate methods of data processing for training a target domain classifier to label text segments. The method includes identifying a set of common keywords with same label from a set of source keywords and a set of target keywords. The method includes training a first classifier, based on the set of common keywords, to label a first set of target text segments. The method includes training a second classifier based on at least a subset of the labeled first set of target text segments. The method includes training a third classifier, based on the first classifier and the second classifier, to label a second set of target text segments, wherein a subset of the labeled second set of target text segments is utilized for re-training the second classifier. The method further includes determining labels of another plurality of target text segments based on the re-trained second classifier.