01. Executive Overview
Social media content on Twitter is unstructured, contains heavy slang, contractions, and noisy punctuation. The goal of this project was to build a clean preprocessing and feature engineering pipeline to accurately categorize tweets into Positive, Neutral, or Negative sentiment buckets using Logistic Regression.
02. Model Performance & Precision Matrix
The model achieved ~70% test accuracy across the three sentiment class labels. Below is the multi-class classification precision breakdown:
Class Wise Precision Breakdown (%)
[REAL PROJECT DATA REQUIRED] Evaluated metric ~70% accuracy per source specification.
03. Preprocessing & Feature Pipeline
The NLP pipeline executes the following stages:
- Regex cleaning (removing URLs, @user handles, hashtags, and special characters)
- Lowercasing & stop-word removal using NLTK
- POS tagging & WordNet lemmatization
- TF-IDF vectorization ($N$-gram range $(1, 2)$, max features $5,000$)
04. Interactive Deployment
A lightweight Streamlit application enables live inference — allowing users to enter custom text strings or tweets and instantly view predicted sentiment class probabilities.