text-emotion-classification

A text emotion recognition application that can be quickly deployed and used locally. You can perform interactive inference simply by running main.py.

中文版

Features

• Local Inference: Loads the sentiment_roberta model directory within the repository for text emotion classification.
• Label Mapping: Reads id -> Chinese Emotion Name mapping from text-emotion.yaml.
• Interactive CLI: Enter text in the command line to output the emotion category and confidence level.

Directory Structure (Key Files)

• main.py: Entry script (run directly).
• sentiment_roberta/: Exported Transformers model directory (contains config.json, model.safetensors, tokenizer, etc.).
• text-emotion.yaml: Label mapping file.
• release-note.md: Release notes (used by GitHub Actions as the release body).

Environment Requirements

• Python 3.10 (Recommended, matches the author's environment; 3.9+ is theoretically compatible but not fully verified).
• Dependency Management: Conda environment (recommended) or venv.
• PyTorch:
  • CPU Inference: Install the CPU version of torch.
  • GPU Inference: Requires an NVIDIA GPU + corresponding CUDA version (the author's environment uses torch==2.10.0+cu128 / torchvision==0.25.0+cu128 built with CUDA 12.8).

The author's conda environment export file is provided: environment.yml.

Installation

Using Conda Environment File (Recommended)

conda env create -f environment.yml
conda activate text-emotion-classification

Usage

python main.py

Follow the prompts to enter text:

• Enter any text: Outputs emotion prediction and confidence.
• Empty input (Enter): Exits the program.

FAQ

• Cannot find model directory sentiment_roberta
  • Ensure sentiment_roberta/ exists in the root directory and contains files like config.json and model.safetensors.
• Inference Device
  • The program automatically selects cuda if available; otherwise, it defaults to cpu.

License

See Apache 2.0 License.