https://github.com/agn-7/emotion-detection-rnn

A bidirectional GRU model to detect discrete emotions in tweets.
https://github.com/agn-7/emotion-detection-rnn

colab colaboratory deep-learning deep-reinforcement-learning emotion-detection google-colab gru jupyter-notebook nlp python python3 rnn sentimental-analysis tensorflow twitter

Last synced: 7 months ago
JSON representation

A bidirectional GRU model to detect discrete emotions in tweets.

• Host: GitHub
• URL: https://github.com/agn-7/emotion-detection-rnn
• Owner: agn-7
• Created: 2020-01-20T08:39:01.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2020-07-22T17:48:35.000Z (about 5 years ago)
• Last Synced: 2025-01-16T13:58:41.493Z (9 months ago)
• Topics: colab, colaboratory, deep-learning, deep-reinforcement-learning, emotion-detection, google-colab, gru, jupyter-notebook, nlp, python, python3, rnn, sentimental-analysis, tensorflow, twitter
• Language: Jupyter Notebook
• Size: 13 MB
• Stars: 2
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Emotion Detection Model

This is the code for training an emotion detection model using GRU presented in:

![GRU Model](model.png "GRU Model")

___

## Train your own model

Requirements (tested with):

- *Python 3.6*

- *Numpy 1.14.5*

- *Pandas 0.24.1*

- *Sklearn 0.19.0*

- *Tensorflow 2.0.0-beta1*

### Installation:

    pip install -r requirements.txt

### To run:

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

### Google-Colab

You can use `handler.ipynb` in order to use in google-colab or jupyter notebook.



---

## Testing

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`. 

## Citation

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}

}