huggingface/torchMoji

Built on 1.2 billion tweets with emoji supervision, the model learns emotional representations transferable to downstream tasks via fine-tuning. The architecture extracts 2304-dimensional emotional feature vectors from text and predicts emoji distributions, enabling both direct emoji classification and transfer learning for sentiment/sarcasm detection on custom datasets. Integrates with scikit-learn and the emoji library for preprocessing and analysis workflows.

922 stars. No commits in the last 6 months.