SOCIAL MEDIA SENTIMENT ANALYSIS & CLASSIFICATION

1. Project Overview & Objective

"My project is a comprehensive Social Media Sentiment Analysis and Classification system. The core objective was twofold: first, to perform an in-depth Exploratory Data Analysis (EDA) on social media posts to understand user sentiment trends across various platforms, countries, and times; and second, to build and evaluate multiple Machine Learning classification models to accurately predict the sentiment (Positive, Negative, Neutral) of new, unseen text data."

=>The complete project workflow is contained within the main.ipynb Jupyter Notebook

2. Technical Stack & Libraries

This project was developed entirely in Python and utilizes a standard, robust set of libraries for Data Science and NLP: Data Manipulation: Pandas, NumPy Text Processing: NLTK and re for Tokenization, Stopword Removal, Stemming (PorterStemmer), and Regex for text cleaning. Visualization: Matplotlib, Seaborn, WordCloud for Creating clear visualizations for EDA (sentiment distribution, temporal trends). Machine Learning: Scikit-learn (sklearn) for Model selection, training, evaluation, cross-validation, and hyperparameter tuning (GridSearchCV).

3. Methodology: The 4-Stage NLP Pipeline

I followed a structured, four-stage pipeline, which is a common best practice in NLP projects.

A. Data Preprocessing & Cleaning

Initial Cleaning: Removed noise from the text data (e.g., URLs, HTML tags, special characters) and standardized the text to lowercase.

Text Transformation: Applied core Natural Language Processing (NLP) techniques:

Tokenization: Breaking text into individual words.

Stopword Removal: Eliminating common, uninformative words (e.g., 'the', 'a', 'is').

Stemming: Reducing words to their root form (e.g., 'running' to 'run') using the Porter Stemmer.

Feature Vectorization: Converted the cleaned text into a numerical format using TF-IDF Vectorization (Term Frequency-Inverse Document Frequency) to be processed by ML models.

B. Exploratory Data Analysis (EDA)

Sentiment Distribution: Visualized the overall counts of Positive, Negative, and Neutral posts.

Temporal Analysis: Analyzed post frequency to determine the Peak Posting Hour and sentiment trends over time.

Platform & Country Analysis: Determined which platforms and countries contribute most to each sentiment type.

Engagement Analysis: Calculated the correlation between Likes and Retweets.

C. Model Training & Selection

Models Evaluated: The performance of three classic, yet powerful, classification algorithms was compared:

1. Logistic Regression

2. Random Forest Classifier

3. Support Vector Classifier (SVC)

Evaluation: Employed Cross-Validation and Grid Search to ensure robust performance, prevent overfitting, and optimize the best model's hyperparameters.

4. Key Results & Insights

The final summary provided several key business and technical insights, which are critical for any stakeholder interested in social media strategy:

Best Model: Highest-performing classification algorithm based on accuracy and F1-Score. Model Accuracy: Quantified performance of the best model (e.g., ~90% Accuracy). Dominant Sentiment: The most frequent overall sentiment in the dataset. Peak Hour: The hour of the day when posting activity is highest. Best Platform: The platform with the highest engagement or most positive sentiment. Likes/Retweets Correlation: The linear relationship between likes and retweets, indicating engagement consistency.