Text Data Augmentation via Semantic Consistency Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data augmentation methods for natural language processing are inadequate due to issues with randomly generated strings not forming sentences and words with similar pronunciations having different meanings, which complicates the application of techniques from voice and image data augmentation.
Innovation Solution
A method and apparatus for augmenting textual data using a data augmenter that generates augmented data based on a chosen data augmentation scheme specific to natural language processing tasks, with a classifier to determine whether the augmented data maintains label information and a consistency determinator to decide on using the augmented data based on classification criteria, including analysis methods for grammar and label comparison.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data augmentation is applied using random string generation methods from voice and image fields, then the quantity of training data increases, but the quality deteriorates because randomly listed strings do not form valid sentences and words with similar pronunciation have different meanings
Solution Approach 1:
The patent changes the parameters of data augmentation by transitioning from random string generation to NLP-specific techniques such as synonym replacement, back-translation, and contextual insertion. These methods maintain linguistic parameters (grammar, semantics, context) while increasing data quantity, thereby resolving the contradiction between quantity and quality.
Solution Approach 2:
The patent introduces NLP-specific processing mechanisms as intermediaries between the raw text and augmented data. These intermediaries (language models, translation systems, synonym databases) ensure that augmented text maintains semantic coherence and grammatical correctness, preventing the quality deterioration that would occur with direct random generation.
2Reliability
If multiple data classification criteria are used to ensure data quality, then the reliability of augmented data improves, but the device complexity increases due to multiple analyzers and classification steps
Solution Approach 1:
The patent segments the data classification process into distinct analyzers, each responsible for specific criteria (grammar checking, semantic consistency, context relevance). This segmentation allows for modular design where each component can be independently optimized and maintained, managing complexity through structured division of labor.
Solution Approach 2:
The classification system performs self-validation through automated consistency determination, where the augmented data is evaluated against predefined NLP task requirements and label information. This self-service mechanism reduces the need for external manual verification while maintaining high reliability standards.
Data Source
AI summary
An apparatus for augmenting textual data according to an embodiment includes a data augmenter configured to generate augmented data by augmenting input textual data according to a data augmentation scheme decided based on a type of natural language processing task of the input textual data and a data classifier configured to classify the augmented data into a positive sample or a negative sample by determining whether or not the augmented data maintains label information of the input textual data based on one or more data classification criteria.