Sentiment Analysis of Product Reviews

Sentiment Analysis of Product Reviews

Online discourse platforms struggle to scale human content moderation to match the volume of user-generated posts. This natural language processing system automates the detection of hate speech, harassment, and discriminatory rhetoric through a fine-tuned transformer architecture. The model is pretrained on broad internet text and subsequently adapted to a labeled corpus of moderated comments using supervised fine-tuning with class-weighted loss to address label imbalance. Beyond binary classification, the system performs fine-grained taxonomy tagging—identifying targeted protected characteristics and severity levels—to enable tiered response workflows. Explainability modules highlight n-gram and attention-weight evidence supporting each moderation decision, facilitating human reviewer oversight and regulatory compliance.

Components

Python 3.8+
NLTK/VADER
Scikit-learn
Flask + Chart.js
BeautifulSoup

Components	Hexkart	Flipkart
Python 3.8+
NLTK/VADER
Scikit-learn
Flask + Chart.js
BeautifulSoup

Note: The components listed are software tools and frameworks required for development.

NLP Preprocessing

Text preprocessing is the foundation of accurate sentiment analysis. The pipeline first cleans raw review text by converting to lowercase, removing HTML tags, special characters, and URLs. Tokenization splits text into individual words. Stop words (common words without sentiment value) are filtered out. Lemmatization reduces words to their dictionary form. BeautifulSoup handles HTML content cleaning when reviews are scraped from e-commerce sites.

VADER Sentiment Analysis

VADER (Valence Aware Dictionary and sEntiment Reasoner) is a lexicon and rule-based sentiment analysis tool specifically tuned for social media and review text. Unlike simple word-counting approaches, VADER understands punctuation emphasis (!!!), capitalization (GREAT vs great), degree modifiers (very, extremely), and negation (not good). It outputs compound polarity scores from -1 (most negative) to +1 (most positive), enabling fine-grained classification.

Visualization Dashboard

Chart.js creates an interactive visualization dashboard within the Flask web application.
Key visualizations:

• Pie chart showing sentiment distribution (positive, negative, neutral).
• Bar chart of rating distributions correlated with sentiment scores.
• Word cloud highlighting most frequent positive and negative keywords.
• Time-series line graph tracking sentiment trends over review dates.

System Architecture

Key Functionalities

Review Collection

• Web scraping e-commerce product reviews using BeautifulSoup
• CSV/Excel file import for batch review analysis
• Direct text input for single review analysis
• Support for Amazon, Flipkart, and other major platforms

Sentiment Classification

• VADER for lexicon-based sentiment scoring
• ML classifiers (Naive Bayes, Logistic Regression) on TF-IDF features
• Hybrid approach combining VADER and ML for improved accuracy
• Multi-class classification: Positive, Negative, Neutral

Data Visualization

• Interactive pie and bar charts using Chart.js
• Word clouds showing prominent positive and negative terms
• Time-series analysis of sentiment trends
• Export analysis reports as PDF