Natural Language Processing (NLP) | E-commerce Analytics | Brand Intelligence
Hybrid Sentiment Analysis — Fashion & Cosmetics
This project demonstrates how natural language processing and data visualization can work together to decode emotions behind beauty and fashion products.An interactive Streamlit web app that analyzes brand sentiment for fashion and cosmetics reviews using a hybrid approach — blending lexicon-based signals with review metadata for transparent, explainable insights.
My Role
Data & Research Analyst
Duration
3 Months
Tools
SQL, Python, Streamlit, Pandas, Scikit-learn, NLTK, Matplotlib, WordCloud
Link
Overview
Beauty and fashion brands receive thousands of product reviews daily, but struggle to derive consistent sentiment insights from them.
Off-the-shelf sentiment models are either too generic or opaque — providing limited transparency into why a product or brand performs well.
The challenge was to design an explainable, hybrid sentiment system that blends linguistic cues with review metadata to produce actionable, brand-level insights.
Developed an interactive Streamlit web app that performs hybrid sentiment analysis on fashion and cosmetics reviews by combining:
Lexicon-based signals (positive/negative keywords, intensifiers, negations)
Metadata features (ratings, verified flag, brand, price)
The app auto-detects key fields, labels reviews with hybrid sentiment categories, and visualizes brand trends, keyword dominance, and product-level performance — all within a recruiter-friendly dashboard UI.
Users can explore the demo dataset in Fast Mode or upload their own CSV for instant analysis.
Research
Analyzed large-scale datasets from Amazon Fashion (800K) and Cosmetics & Beauty Reviews (Kaggle) to understand user tone and feedback structure.
Findings from linguistic and exploratory analysis guided the app’s feature engineering:
Reviews often mix emotional tone with practical attributes (e.g., price, packaging, shade match).
Ratings alone fail to capture sarcasm or nuanced satisfaction.
Combining lexicon rules + numeric metadata yields more stable and explainable sentiment classifications.
The hybrid model emerged as a balance between accuracy and interpretability.
Design
Built a Streamlit dashboard with elegant, pastel-themed visuals inspired by fashion and cosmetics aesthetics.
Included modules for keyword analysis, brand leaderboard, price vs sentiment, and trend tracking.
Integrated interactive filters (brand, verified status, price range) for user exploration.
Added CSV export, word clouds, and monthly trend visualization for comprehensive reporting.
Deployed the app on Streamlit Cloud in Fast Mode for seamless recruiter demos.
Results
Enabled fast, transparent sentiment scoring across thousands of fashion and beauty reviews.
Improved explainability and interpretability vs standard black-box models.
Revealed brand-level sentiment patterns and customer pain points with high precision.
Reduced analysis time by 80% through automation and intuitive visualizations.
Deployed publicly for users and brand managers to test live.
🌟 Takeaway
Hybrid Sentiment Analysis bridges fashion intuition with AI transparency — blending lexicon-based interpretability and machine learning simplicity.
It demonstrates how data storytelling and brand analytics can combine to reveal emotional drivers behind customer satisfaction in the beauty and fashion industry.
