Token String Augmentation for Stable Tag Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Documents from infrastructure businesses are difficult to understand without specialized knowledge, making it challenging to prepare high-quality training data for extracting expressions related to troubles, and existing data augmentation techniques introduce noise or are ineffective for text data.
Innovation Solution
An information processing apparatus that augments token strings by looping and shifting the start position, generating multiple augmented token strings while maintaining the original arrangement, and uses a tag estimation model to determine tags through voting, reducing noise and improving model stability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data augmentation techniques (random insertion, deletion, or exchange of words) are applied to document data, then the quantity of training data is increased, but noise is introduced and estimation stability deteriorates
Solution Approach 1:
The patent creates augmented token strings by copying and looping the original token string multiple times, generating multiple versions that preserve the original information without introducing noise. This allows increasing training data quantity while maintaining data quality and estimation stability.
2Quantity of substance
If simple data augmentation methods (enlargement, reduction, rotation, trimming) are applied to text data, then data quantity is increased, but the methods are ineffective because they cannot be applied to continuous values representing characters or words
Solution Approach 1:
The patent changes the parameter representation by using discrete token strings instead of continuous embedding values for data augmentation. This allows applying augmentation operations to text data in a manner analogous to image processing, making the augmentation process both effective and easy to implement.
3Quantity of substance
If words in sentences are randomly inserted, deleted, or exchanged to augment data, then training data quantity increases, but the meaning of sentences may change and noise is introduced
Solution Approach 1:
The patent creates augmented token strings by copying and looping the original token string multiple times, generating multiple versions that preserve the original information without introducing noise. This allows increasing training data quantity while maintaining data quality and estimation stability.
4Adaptability or versatility
If synonyms are used to replace specific words for data augmentation, then data variety is increased, but it is difficult to prepare appropriate synonyms
Solution Approach 1:
The patent changes the parameter representation by using discrete token strings instead of continuous embedding values for data augmentation. This allows applying augmentation operations to text data in a manner analogous to image processing, making the augmentation process both effective and easy to implement.
Data Source
AI summary
According to one embodiment, an information processing apparatus includes a processor including a hardware. The processor performs augmentation on a token string included in acquired document data so as to maintain an arrangement of an original token string to generate a plurality of augmented token strings. The processor estimates a tag to be appended to each of the augmented token strings. The processor determines a tag to be appended to the token string based on the tag estimated for each of the augmented token strings.