Analyze and detect emotions in text with AI-powered precision using the RoBERTa-based GoEmotions model.
The RoBERTa-base GoEmotions model was fine-tuned on the GoEmotions dataset, which consists of Reddit comments labeled with 28 different emotions. Since GoEmotions is a multi-label dataset, the model was trained to predict one or more emotions for each input text. The base model, RoBERTa-base, is a pre-trained transformer-based language model known for its ability to capture contextual meaning from text efficiently. The training process involved multi-label classification, where the model learned to assign probabilities to multiple emotions per input, with a typical classification threshold set at 0.5.
To enhance inference speed and efficiency, an ONNX version of the model was created, including an INT8 quantized variant. This conversion allows for faster predictions, smaller dependencies, and cross-platform compatibility. The quantized model further reduces the file size by 75% while maintaining near-original accuracy, making it ideal for applications where efficiency and portability are crucial.
Enter a sentence to detect emotions like joy, sadness, or anger.
| Emotion | Score |
|---|