Emotion is one of the basic instincts of a human being. Emotion detection plays a vital role in the field of textual analysis. At present, people’s expressions and emotional states have turned into the leading topic for research works.In this project, Our primary goal is to detect human’s emotion from text input through some Deep Learning Model.
Project is created with:
- Google Colab
Language used:
- Python
Library used:
- Pytorch
- NLTK
- Tensorflow
- sklearn
Word Emebedding:
Here tweet emotions from SemEval-2018 Affect in Tweets Distant Supervision Corpus (AIT-2018 Dataset) is used.This dataset has two columns content and sentiment. It has 20000 unique text classified with 6 emotions such as anger, love, surprise, fear, joy, sadness. This dataset contains lots of emotions but they mainly clustered those emotions into 4 basic emotions such as anger, fear, joy, sadness.
- Convolutional Neural Network (CNN)
- Bidirectional Long Short Term Memory (Bi-LSTM)
Accuracy of the two models - Convolutional Neural Network (CNN) and Bidirectional Long Short Term Memory (Bi-LSTM)
are 90.00% and 92.33% respectively.
In this project, we have used ”tweet emotions from SemEval-2018 Affect in Tweets Distant Supervision Corpus (AIT-2018 Dataset)”. As it’s a noisy dataset, first we had to do the preprocessing such as punctuation remove and converted emojis into texts, tokenization, contraction. Then we have implemented two models and compared them. After comparing them, we observed that Bi-directional long short term memory model gave the highest accuracy which is 92.33%. After completion of our project, we observed that our project can detect emotions from texts.