Emotion Text Detection Project

This project is of the course Computational Linguistics Team Laboratory for the MSc in Computational Linguistics at the University of Stuttgart.

Description

Emotion text detection involves classifying text based on emotional content, a critical and challenging aspect of Natural Language Processing (NLP) due to the nuanced nature of human emotions. In this project, we aimed to perform a multilabel classification on the ISEAR dataset (CISA, University of Geneva, 2024-07-16), which includes 7 classes - Joy, Sadness, Disgust, Guilt, Shame, Fear, and Anger. It consists of textual descriptions annotated with these emotion labels. The primary research question we aimed to address is: "How can different machine learning models and embedding techniques, which capture contextual information, be utilized to effectively classify fine-grained emotional content in text?"

Results from experiments

Setup

Create the virtual environment:-

Please use Python version 3.12.3 to be compatible with the libraries used in the project

python -m venv venv
source venv/bin/activate

Install the project and the dependencies using:-

pip install .

Make sure that the datasets are present in a folder named datasets at the root level.

|-- config
|-- scripts
|-- dataanalysis
|-- datasets

Ensure that you have word embeddings downloaded to run the GloVe based Seq. Neural Networks & LSTM GloVe in the below locations.

Embeddings were downloaded from this website: https://nlp.stanford.edu/projects/glove/ , 'glove.6B.zip'

Locations:

Seq. NNs: scripts/advanced_classifier/word_embeddings/glove.6B/glove.6B.300d.txt

LSTM GloVe: scripts/advanced_classifier/lstm/glove.6B.100d.txt

To visualize the performance of various models for each metric, execute the following command:

Note: A similar visualization process has been applied for the sequence neural network models as well.

python3 -m scripts.advanced_classifier.lstm.lstm_model_comparison_visualizer

Run

python ./scripts/main.py