This is the code for training an emotion detection model using GRU presented in:
Requirements (tested with):
- Python 3.6
- Numpy 1.14.5
- Pandas 0.24.1
- Sklearn 0.19.0
- Tensorflow 2.0.0-beta1
pip install -r requirements.txt
After downloading/cloning, put the dataset in the data folder.
To use the dataset in the paper you can download tweets based on their tweet ids available with their classes in ./data/ and remove the hashtags at the end of each tweets. The final dataset should have the following format: id, text, emotion with one record (tweet) per line.
The embedding file should be placed in ./vectorss/
Use the configuration.cfg to set the name of dataset and embedding file, maximun numer in the vocabulary (max_features), maximum length of terms in the text (maxlen), bactch size and number of epochs to run the training.
Then run the handler.py:
python handler.py
You can use handler.ipynb in order to use in google-colab or jupyter notebook.
You can download the trained models used for the paper at: https://drive.google.com/open?id=1TXEbHMTA_AWPFC8bbt7WiBtfT3jVy8cG .
To run, put the test file into the data forlder. test file should be one tweet per line with no additional columns. set the name of the file in test_configuration.cfg and run handler-test.py.
Please use the following citation when using the code or the paper:
@article{seyeditabari2019emotion, title={Emotion Detection in Text: Focusing on Latent Representation}, author={Seyeditabari, Armin and Tabari, Narges and Gholizadeh, Shafie and Zadrozny, Wlodek}, journal={arXiv preprint arXiv:1907.09369}, year={2019} }