Synthetic Complaint Email Generation for Class-Imbalanced Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems struggle with class imbalance in identifying complaint emails due to the scarcity of complaint data, leading to inaccurate classification and increased false positives, which is time-consuming and costly.
Innovation Solution
A platform and language agnostic synthetic text generation module is employed to generate synthetic complaint emails by converting non-complaint emails into complaint-like emails using text style transfer and generation models, augmenting the labeled data to improve classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If keyword-based filters are used to identify complaint emails, then manual work is reduced, but false positives increase significantly
Solution Approach 1:
The patent introduces an intermediary classification system that processes emails between automatic keyword filtering and manual review. This intermediary layer uses machine learning models to score and rank emails, filtering out obvious non-complaints while preserving potential complaints for human review, thereby reducing false positives while maintaining productivity
Solution Approach 2:
The system performs preliminary classification and scoring of emails using trained models before manual review. By pre-identifying and filtering out non-complaint emails through automated analysis of email content, tone, and context, the system reduces the burden on manual reviewers and decreases false positives in the remaining set
2Measurement precision
If manual complaint identification is performed, then accuracy is maintained, but time and cost increase significantly
Solution Approach 1:
The system applies partial automation by using machine learning models to handle the majority of email classification tasks, particularly those with clear indicators. Manual review is reserved only for borderline cases or emails requiring human judgment, thus achieving high accuracy while significantly reducing overall manual review time
Solution Approach 2:
The patent segments the email classification process into multiple stages: initial automated filtering, intermediate scoring and ranking, and final manual review for uncertain cases. This segmentation allows the system to leverage automated efficiency for clear cases while maintaining human accuracy for complex cases, optimizing both time and precision
3Quantity of substance
If labelled data distribution is highly skewed due to scarcity of complaint emails, then data availability is maintained, but classification system performance deteriorates
Solution Approach 1:
The patent employs synthetic data generation techniques to create artificial complaint email samples that mimic real complaint characteristics. These synthetic copies are generated using techniques such as back-translation, paraphrasing, and data augmentation, providing additional training data to balance the skewed distribution and improve model performance without requiring more real complaint data
Solution Approach 2:
The system applies parameter changes to existing complaint data through various transformations including synonym replacement, sentence restructuring, and contextual variations. These parameter changes create diverse versions of complaint emails, effectively increasing the quantity and variability of training data while maintaining the underlying complaint semantics
Data Source
AI summary
A computer based system and method for synthetic text generation includes a processor. The processor implements a text style transfer algorithm to first input data to generate complaints data from non-complaint emails data associated with a plurality non-complaint emails. The processor converts the plurality of non-complaint emails into a first set of complaint emails based on implementing the text style transfer algorithm and implements a text generation model algorithm to second input data to generate a second set of complaint emails from a plurality of complaint emails. The processor also generates a set of synthetic complaint emails based on the generated first set of complaint emails and the second set of complaint emails; trains a model based on the generated synthetic complaint emails; and applies the trained model to a new set of emails to resolve class imbalance in automatic complaint identification from the new set of emails.