🐦 Tweet Sentiment Analyzer

A Streamlit web application to predict the sentiment of tweets. This project uses a Logistic Regression model with TF-IDF features to classify tweets into Positive, Neutral, or Negative categories. It also displays the probabilities for each class in an interactive bar chart.
Data link : https://www.kaggle.com/datasets/jp797498e/twitter-entity-sentiment-analysis

---

🔹 Features

• Preprocess tweets: remove stopwords, URLs, mentions, hashtags, and punctuation
• Lemmatization for better text normalization
• TF-IDF vectorizer for feature extraction
• Logistic Regression model trained on Twitter dataset
• Color-coded sentiment output in Streamlit
• Easy-to-use web interface

---

📂 Repository Contents

File	Description
app.py	Streamlit web application
logistic_model.pkl	Trained Logistic Regression model
tfidf_vectorizer.pkl	Pickled TF-IDF vectorizer
requirement.txt	Python dependencies
twitter_training.csv	Training dataset
sentiment_analysis.ipynb	Notebook with data preprocessing and model training
README.md	Project description and instructions

---

💻 Installation

1. Clone the repository

git clone https://github.com/MohamedAli1937/Sentiment-Analysis-Web-App.git

⚙️ Install dependencies

pip install -r requirements.txt

🎮 Run the Streamlit app

streamlit run app.py

🤔 Prediction Function

def predict_sentiment(text):
    clean_text = clean_tweet_stopword_lemmatize(text)  # your cleaning + lemmatization function
    vectorized = vectorizer.transform([clean_text])
    return lr_model.predict(vectorized)[0]

---

🧠 How It Works

1. Preprocessing:

• Lowercasing, removing URLs, mentions, hashtags, punctuation
• Stopwords removal and lemmatization

2. Feature Extraction:

• TF-IDF converts text to numerical vectors

3. Model:

• Logistic Regression predicts sentiment class

4. Output:

• Sentiment class (Positive/Neutral/Negative)

---

🚀 Future Improvements

1. Better Models

• Use DistilBERT or RoBERTa for more accurate predictions
• Deep learning models capture context better than Logistic Regression

2. Emotion Detection

• Expand beyond Positive/Neutral/Negative
• Detect specific emotions: Happy, Sad, Angry, Fear, Surprise, etc.
• Use libraries like NRCLex or train multi-class classifiers

3. Data Enhancements

• Add more neutral tweets to improve model balance
• Include tweets in multiple languages

4. UI/UX Improvements

• Show word clouds for positive/negative words
• Display historical sentiment trends from multiple tweets
• Add interactive charts for probabilities

5. Deployment

• Deploy online via Streamlit Cloud, Heroku, or AWS
• Make a public demo for users to try

---

⚠️ Known Limitations

• The current Logistic Regression model sometimes misclassifies neutral tweets as positive or negative.

• This happens because the training dataset has fewer neutral examples, making the model biased toward positive/negative sentiment.

• Probabilities for neutral predictions may be less reliable compared to positive or negative.