Sentiment analysis is a A natural language processing technique used to determine the data belongs to which class/type of emotions. Sentiment analysis is often performed on textual data to help businesses monitor brand and product sentiment in customer feedback and understand customer needs.

Dataset:

The dataset contains 20,000 tweets that were collected through API for classifying emotions.

It contains 6 classes of emotions:

• Anger
• joy
• fear
• love
• surprise
• sadness

The dataset can be found at:

• https://huggingface.co/datasets/dair-ai/emotion

or

• https://paperswithcode.com/dataset/emotion

The Pipeline:

• 1 - Text Preprocessing Dataset Cleaning: Check for null values Check for duplicated values Remove unwanted patterns as: #, &, punctuation, ... etc.

• 2 - Text Normalization Used NLTK Library and includes 3 Techniques: Tokenization: Converting sequence of texts into smaller parts. Slemming: Reducing a word to its word stem (root). Lemmatization: Reduce the word better to its root word, or lemme, good.

• 3 - Word Embedding BOW (Bag Of Words): Accuracy 80% TF-IDF: Accuracy 77% Word2Vec with LSTM: 64%

• 4 - Modeling & Evaluation