Virtual Assistant Response Harmonization via Confidence Score Normalization

The patent applies parameter changes by transforming confidence scores from different scales into a normalized common scale. Each assistant's confidence score is adjusted using learned normalization parameters specific to their domain expertise, enabling direct comparison while preserving the original scale variations that reflect domain-specific confidence levels.

The patent introduces an intermediary normalization layer between the multiple virtual assistants and the final response selection. This intermediary component learns domain-specific normalization parameters that mediate the comparison process, allowing confidence scores from different assistants to be fairly evaluated without directly comparing raw scores on different scales.

The patent applies preliminary action by pre-learning domain-specific weights and normalization parameters during a training phase before actual query processing. This preliminary learning phase creates a knowledge base that simplifies the runtime operation, where the system only needs to apply pre-determined weights rather than calculating them in real-time, thus reducing operational complexity while maintaining high accuracy.

The system performs self-service by automatically learning and adapting domain-specific weights through the training process without requiring manual configuration. The normalization parameters and domain weights are derived automatically from training data, allowing the system to self-optimize its response selection mechanism while managing its own complexity through automated learning rather than manual setup.