State-of-the-art research achieves a model in emotion prediction with 17% top-1 / 44% top-5 accuracy.

ChipBrain nearly doubles SOTA with a model that predictions customer emotions with 31% top-1 / 64% top-3 accuracy. (by fine-tuning models for specific contexts, like sales and customer service, and solving a hierarchy over classes based on noise estimation from confident learning).

ChipBrain re-architects models using robust pre-training methods combined with modern transfomer-based methods, now training our models from scratch without dependencies on prior emotion detection techniques, and significantly increasing accuracy to 38% top-1 / 70% top-3 (over 90% when thresholding via predicted probabilities)