Verbal Harassment Detection Training via Synthetic Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning models for verbal harassment detection face challenges in obtaining sufficient high-quality training data, particularly due to imbalanced datasets and noisy labeling, which affects their accuracy and efficiency in ride-sharing settings.
Innovation Solution
The approach involves generating verbal harassment heuristics using labeled segments, applying these heuristics to unlabeled segments to assign labels, selecting subsets based on likelihood thresholds, and pre-training machine learning models using a combination of labeled and randomly selected segments, followed by updating the models with higher-quality training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are trained using traditionally available training data, then training can proceed with existing resources, but the accuracy and detection performance are limited due to imbalanced and noisy datasets
Solution Approach 1:
The system performs preliminary actions by generating synthetic training data samples that simulate verbal harassment scenarios before actual detection is needed. This pre-generation of balanced training data with proper labeling enables the model to learn from diverse scenarios without waiting for sufficient real-world examples, directly addressing the data imbalance and quality issues
Solution Approach 2:
The system introduces an intermediary synthetic data generation process that bridges the gap between limited real training data and the needs of accurate detection. This intermediary layer creates additional training samples with controlled characteristics, allowing the model to learn from a more balanced and comprehensive dataset without relying solely on scarce real-world examples
2Measurement precision
If more training data is collected to improve model performance, then detection accuracy may improve, but the time required for data collection and model training increases
Solution Approach 1:
The system creates copies of existing training data through synthetic data generation, replicating verbal harassment scenarios with varied parameters. This copying approach allows the model to learn from multiple instances of similar scenarios without requiring extensive manual data collection, significantly reducing the time investment while maintaining detection accuracy
Solution Approach 2:
The system varies parameters in synthetic data generation (such as speech patterns, contextual scenarios, and harassment types) to create diverse training samples efficiently. By changing parameters programmatically rather than collecting diverse real-world data manually, the system achieves comprehensive training coverage with reduced time investment
3Measurement precision
If manual labeling of training data is performed to ensure high quality, then data accuracy improves, but the complexity and resource requirements increase
Solution Approach 1:
The system implements self-service labeling through automated synthesis and labeling of training data. Rather than requiring manual expert labeling, the synthetic data generation process automatically creates properly labeled samples based on defined verbal harassment scenarios and speech patterns, eliminating the need for complex manual labeling processes while maintaining high label accuracy
Data Source
AI summary
In some cases, lower quality, large scale training data can be automatically generated by automatic labeling. The generated training data can be used to pre-train a machine learning model. For instance, the model can be a model for detection of verbal harassment. Parameters of the pre-trained model can be refined or updated using another one or more higher-quality sets of training data, with which the model can be subsequently trained.