Dual-Prediction Model for Probabilistic Data Labeling

The patent introduces a probabilistic label assignment mechanism that acts as an intermediary between the defined source of truth and complex tasks without definitive tests. Instead of requiring direct definitive labeling, the system uses probability thresholds and multiple prediction models to assign labels, enabling handling of complex tasks while maintaining labeling accuracy through the intermediary probabilistic layer.

The system changes the parameter from binary definitive labels to probabilistic labels with confidence scores. By transforming the labeling output from certain/uncertain to probability-based, the system can handle complex tasks where definitive tests are unavailable while maintaining accuracy through statistical confidence thresholds and model predictions.

The patent applies partial action by using probability thresholds rather than requiring definitive classification. Candidates are included in the pool if they exceed a probability threshold (e.g., >0.5), rather than requiring definitive proof. This partial inclusion criterion maintains labeling purity by filtering out clearly incorrect candidates while increasing the candidate pool size by including those with sufficient probabilistic evidence.

The system incorporates feedback through iterative model training where predicted probabilities are used to refine future predictions. The dual-prediction model system uses feedback from historical data and previous predictions to continuously improve accuracy, allowing the system to confidently include candidates with probabilistic labels while maintaining high purity through feedback-driven model refinement.

The patent segments the prediction system into two distinct prediction models: a first prediction model for initial labeling and a second prediction model for refinement. This segmentation allows each model to specialize in specific aspects of prediction, improving overall accuracy while managing complexity through modular architecture where each model can be independently trained and optimized.

The system adds a probabilistic dimension to the prediction output, transforming binary classifications into probability distributions. By operating in the probability space dimension rather than just the binary classification dimension, the system achieves higher prediction accuracy while the added complexity is managed through standardized probability thresholding and calibration techniques.