Chatbot Training Data Outlier Adjustment by Nearest-Neighbor Labels

The patent applies preliminary action by performing outlier detection and adjustment before the actual model training process. The system identifies and adjusts outlier data points in advance, creating a cleaned training dataset that prevents outliers from compromising model accuracy during training. This proactive approach ensures high model accuracy while containing processing complexity to a predefined preparation phase rather than ongoing training.

The patent segments the data processing workflow into distinct phases: outlier detection, outlier adjustment, and model training. By separating the outlier handling process from the training process, the system manages complexity through modular processing steps. The outlier detection and adjustment operations are performed as a preliminary dataset preparation stage, independent from the subsequent training phase, thereby improving overall system manageability.

The system performs outlier detection and data cleaning as a preliminary action before model training begins. By completing the time-consuming outlier adjustment process in advance during dataset preparation, the actual model training can proceed more efficiently without interruptions. This separates the processing time into a one-time preparation phase rather than continuously impacting training time.

The patent implements self-service mechanisms where the system automatically detects, identifies, and adjusts outlier data points without requiring manual intervention. The automated outlier detection algorithms independently analyze the training data, identify outliers based on predefined criteria, and perform adjustments, thereby reducing the time that would otherwise be spent on manual data cleaning and verification processes.