Automated Labeling for Text Classification Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning models in natural language processing require extensive labeled training samples, which are time-consuming and costly to obtain manually, and existing models struggle to meet application requirements in vertical fields due to differences between general and specific text data, resulting in suboptimal classification performance.
Innovation Solution
An information processing device and method that constructs sample unit sets based on minimum labeled sample units, generating labeled training samples automatically, and combines a pre-trained word embedding extraction model with a contextual representation model to improve text classification performance in specific fields.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling is used to obtain training samples, then the quality and accuracy of labeled samples is improved, but the time cost and economic cost increase significantly
Solution Approach 1:
The system performs preliminary automatic labeling using pre-trained models to generate initial labels for training samples. This preliminary action reduces the burden of manual labeling while maintaining acceptable accuracy, allowing manual reviewers to focus only on correcting or verifying critical samples rather than labeling everything from scratch.
Solution Approach 2:
A semi-automatic labeling system acts as an intermediary between completely automatic and completely manual labeling. The system uses pre-trained models to generate initial labels, which are then refined through minimal manual intervention. This intermediary approach balances automation efficiency with human oversight quality, significantly reducing time cost while preserving labeling accuracy.
2Adaptability or versatility
If deep learning models are trained using general universal texts, then the model can learn general language patterns, but the classification performance in specific vertical fields deteriorates due to domain differences
Solution Approach 1:
The system applies local quality by using domain-specific pre-trained models for each vertical field (e.g., medical, legal, financial) rather than a single general model. Each model is specialized to capture the unique language patterns, terminology, and contextual nuances of its specific domain, thereby improving classification performance while maintaining adaptability through the modular architecture that allows switching between domain models.
Solution Approach 2:
The system changes parameters by adapting pre-trained models to specific domains through domain-specific corpus pre-training and fine-tuning. This involves adjusting model parameters and weights based on domain-specific data characteristics, enabling the model to maintain general language understanding while achieving high precision in vertical field classification tasks.
3Measurement precision
If a large number of training samples are required to achieve model performance, then the model performance is improved, but the input far exceeds the output making actual application difficult
Solution Approach 1:
The system performs preliminary actions by pre-training models on large domain-specific corpora before actual application. This pre-training phase captures essential domain knowledge and patterns, allowing the model to achieve high performance with fewer additional training samples needed during deployment, thereby reducing the complexity of sample processing in practical applications.
Solution Approach 2:
The system enables self-service by using automatically generated labels from pre-trained models to create training datasets without requiring extensive manual labeling. The model essentially labels its own training data through self-supervised or weakly-supervised learning approaches, reducing the complexity of data preparation while maintaining model performance.
Data Source
AI summary
An information processing device, an information processing method, a computer readable storage medium are provided. The information processing device comprises processing circuitry configured to: construct, for each of a plurality of indexes, a sample unit set for the index based on a plurality of minimum labeled sample units related to the index which are obtained and labeled from an original sample set; and extract, for at least a part of the constructed plurality of sample unit sets, a minimum labeled sample unit from each sample unit set, and generate a labeled training sample based on the extracted minimum labeled sample unit. A sample unit set is constructed based on minimum labeled sample units that are labeled manually, and a labeled training sample is generated automatically based on such sample unit sets, thereby generating the labeled training sample automatically to a certain degree, and reducing manual participation.