NLI-Based Domain Text Labeling With Few-Shot Fine-Tuning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing text classification approaches, such as sentiment analysis, require large amounts of labeled training data, which is time-consuming and labor-intensive, and pre-trained language models are not accurate for domain-specific text classification tasks without fine-tuning.
Innovation Solution
A method involving a pre-trained natural language inference (NLI) model applied to a set of texts and hypothesis statements, followed by a few-shot learning model to generate a fine-tuned NLI model that labels input text accurately, reducing the need for extensive manual labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised learning approaches are used for text classification, then model accuracy can be improved, but large amounts of labeled training data are required which increases time consumption and human effort
Solution Approach 1:
The patent applies pre-trained NLI models to generate labeled training data before the actual model training process. By performing preliminary labeling using pre-trained models, the system creates a ready-to-use labeled dataset that eliminates the need for time-consuming manual annotation, thus resolving the contradiction between achieving high accuracy and reducing labeling time
Solution Approach 2:
The system uses automated NLI models to label texts independently without requiring extensive human intervention. The pre-trained models self-generate labeled data by analyzing text-hypothesis pairs, enabling the system to serve its own data labeling needs and significantly reducing human effort while maintaining model accuracy
2Productivity
If pre-trained language models are used for domain-specific text classification, then deployment speed is improved, but accuracy for domain-specific tasks deteriorates without fine-tuning
Solution Approach 1:
The patent performs preliminary fine-tuning of pre-trained NLI models using automatically generated labeled data specific to the target domain. This preliminary adaptation prepares the model for domain-specific tasks while maintaining efficient deployment, resolving the contradiction between quick deployment and domain-specific accuracy
Solution Approach 2:
The system adjusts model parameters through fine-tuning on domain-specific labeled data generated by the NLI model. By changing the model's internal parameters to better fit the specific domain characteristics, the system maintains both deployment efficiency and improved domain-specific classification accuracy
Data Source
AI summary
In an embodiment, a set of texts associated with a domain is received. A set of hypothesis statements associated with the domain is received. A pre-trained natural language inference (NLI) model is applied on each of the received set of texts and on each of the received set of hypothesis statements. A second text corpus associated with the domain is generated. The generated second text corpus corresponds to a set of labels associated with the domain. A few-shot learning model is applied on the generated second text corpus to generate a third text corpus associated with the domain. The generated third text corpus is configured to fine-tune the applied pre-trained NLI model, and the fine-tuned NLI model is configured to label an input text associated with the domain. A display of the labelled input text on a display device is controlled.