Conversational Data Augmentation for ML Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing commercial datasets for machine learning models are often non-domain specific or outdated, limiting their effectiveness in training high-quality models.
Innovation Solution
A system and method for automatically augmenting and labeling conversational data using a combination of text-to-text machine learning models, label-generating models, and deep learning entity models, which generate paraphrases, new phrases, and labeled data from limited pre-labeled data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing commercial datasets are used for training machine learning models, then the training process can be initiated, but the quality and effectiveness of the model is limited due to the datasets being non-domain specific or outdated
Solution Approach 1:
The system performs preliminary actions by automatically generating synthetic domain-specific training data before model training begins. Data generation models create realistic conversational datasets tailored to specific domains, and labeling models pre-process this data with accurate annotations, ensuring high-quality training data is ready in advance rather than relying on outdated commercial datasets
Solution Approach 2:
The system enables self-service by using AI models to automatically generate and label training data without human intervention. The data generation models create domain-specific datasets, and the labeling models automatically annotate them with relevant information, allowing the system to serve its own data needs independently rather than relying on external commercial datasets
2Reliability
If more training data is collected to improve model performance, then the model quality improves, but the time and resources required for data collection and labeling increase
Solution Approach 1:
The system replaces manual mechanical processes of data collection and labeling with automated AI models. Instead of human collectors gathering data and human annotators labeling it, data generation models automatically create synthetic training data and labeling models automatically annotate it, substituting the mechanical human labor process with an automated computational system that operates much faster
Solution Approach 2:
The system changes the fundamental parameter of data origin from externally collected real-world data to internally generated synthetic data. By adjusting this parameter, the system can rapidly produce unlimited amounts of domain-specific training data with consistent quality and accurate labels, eliminating the time-consuming external data collection process while maintaining data quality for model performance
Data Source
AI summary
A method implemented via execution of computing instructions configured to run at one or more processors and stored at one or more non-transitory computer-readable media. The method can include generating training data for an intent classification machine learning model by: (a) determining, via a text-to-text machine learning model, one or more respective paraphrases for each sample phrase of training phrases; (b) generating, via a label generating machine learning model, labeled data based on unlabeled live logs by: (i) determining live-log samples from the unlabeled live logs based at least in part on: a respective timestamp of each live log of the unlabeled live logs, or random sampling; and (ii) generating, via the label generating machine learning model, the labeled data based on the live-log samples and one or more labeling functions; and (c) adding the one or more respective paraphrases for the each sample phrase of the training phrases and the labeled data to the training data. In certain embodiments, a respective quantity of the one or more respective paraphrases can vary for the each sample phrase of the training phrases. In some embodiments, the method further can include transmitting the training data, as generated, to the intent classification machine learning model for training. Other embodiments are described.