100-lines-text-emotion-classification

Introduction

A easy-to-build emotion classification pipeline with ktrain, which is a lightweight wrapper for the deep learning library TensorFlow Keras. We used BERT pre-trained model and trained on a unified dataset which consist of 12 emotion corporas.

Motivation

1. ktrain is very friendly for the beginner
2. Emotion datasets are usually built in different ways. This leads to several research gaps: supervised models often only use a limited set of available resources. Thus, we a promising unified emotion corpora from Institut für Maschinelle Sprachverarbeitung, Universität Stuttgart. The authors selected 12 emotion datasets and reannotated them. We expect to build more general model via this dataset.

---

How to use

1. Build virtual environment

python3 -m venv venv
. venv/bin/activate

2. Install required package

pip install -r requirements.txt
(or) pip install -e .

3. Prepare data:

• you can download the unified emotion corpora here
• The default dataset is in json or tsv format.
• Each datapoint has following information: 'ID', 'Corpora', 'Text', 'Emotion'

4. Fine-tune the pre-trained model:

• you can download the model I already fine-tuned here
• make checkpoints directory in the main root and put checkpoint files inside
• alternatively, you can fine-tune you own model

---

Result sample

Input Text: Always love 'Jeni's' ice cream🍨💓#my #favorite #icecream #ohaio #yum #delicious #happy… https://t.co/JtQ9a1Ag1z
Predicted Label --> joy /      Ground Truth Label --> joy

Input Text: @masters_say sounds like a perfect night that I miss spending with you! #imiss12303
Predicted Label --> sadness /   Ground Truth Label --> sadness

Input Text: You can't beat a bit of Division. Interzone.
Predicted Label --> noemo /     Ground Truth Label --> joy

Input Text: @JordanWooten yeah, yeah. we'll see....can't ruin Christmas.
Predicted Label --> surprise /  Ground Truth Label --> surprise

Input Text: Bon. On va tenter la cuisine avec l'huile d'arachide ...
Predicted Label --> fear /      Ground Truth Label --> fear

---

Evaluation

---

To-dos

• Evaluate on individual dataset
• Build similar pipeline for via pytorch-lightning
• More firendly to other dataset

---

Contact

If you have any question or suggestion, feel free to contact me at [email protected]. Contributions are also welcomed. Please open a pull-request or an issue in this repository.

---

Citation

@inproceedings{bostan-klinger-2018-analysis,
    title = "An Analysis of Annotated Corpora for Emotion Classification in Text",
    author = "Bostan, Laura-Ana-Maria  and
      Klinger, Roman",
    booktitle = "Proceedings of the 27th International Conference on Computational Linguistics",
    month = aug,
    year = "2018",
    address = "Santa Fe, New Mexico, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/C18-1179",
}