The Sentimentalizer is a machine learning project that performs sentiment analysis on text data, classifying inputs as positive, negative, or neutral.
Originally developed for a hackathon, the project demonstrates how natural language processing (NLP) and classical machine-learning models can be combined to build a fast, interpretable sentiment classifier.

This project includes a notebook showing the full workflow:
data loading → text preprocessing → vectorization → model training → evaluation → real-time predictions.

• End-to-end sentiment analysis pipeline
• Text cleaning & preprocessing (tokenization, stopwords, lemmatization)
• Feature extraction using TF-IDF
• Multiple sentiment classifiers tested, including:
  • Logistic Regression
  • Naive Bayes
  • Support Vector Machine (SVM)
• Model comparison & evaluation metrics included
• Real-time sentiment prediction examples
• Cleaned dataset (save.csv) included for reproducibility

Text entries are imported from the dataset and undergo preprocessing:

• Lowercasing
• Removing punctuation & special characters
• Tokenization
• Stopword removal
• Lemmatization/stemming

This ensures cleaner, more consistent input for the models.

The preprocessed text is vectorized using TF-IDF, which transforms raw text into numerical feature vectors based on word frequency and importance.

Several classical NLP models are trained and evaluated:

• Logistic Regression
• Multinomial Naive Bayes
• Linear SVM

Models are scored using:

• Accuracy
• Precision
• Recall
• F1 Score
• Confusion Matrix

The notebook includes complete comparisons to help determine the best-performing classifier.

Once trained, the model can classify any text input, e.g.: "That movie was surprisingly good!" → Positive "I expected better, it was disappointing." → Negative "It's okay, nothing special." → Neutral

The notebook includes multiple real-world examples.

While results depend on the dataset used, typical performance from the notebook shows:

• TF-IDF + Linear SVM generally performs best
• Logistic Regression performs competitively
• Naive Bayes provides strong speed and baseline accuracy

The project emphasizes interpretability and ease of use rather than deep learning.

• Classifying customer reviews
• Social media sentiment monitoring
• Product feedback analysis
• Chatbots and automated response systems
• Quick prototyping for NLP hackathons

The hackathon.ipynb file contains:

• Full preprocessing pipeline
• Training and evaluation of multiple models
• Visual performance comparisons
• Interactive prediction cells
• All code required to reproduce the results

Created by Yasmin Modarai and Safwan Hasan for a machine learning hackathon.
Extended, organized, and documented for educational use and fast prototyping.

This project is open for personal and educational use. For commercial use, please attribute the original author.