GlowSense ✨ — Beauty & Skincare Product Intelligence

GlowSense is a premium, portfolio-ready product intelligence dashboard designed for beauty and skincare shoppers. It aggregates real-world customer reviews from Sephora, Ulta, and Amazon, runs custom lexical sentiment analysis, and calculates an explainable multi-factor Glow Score to rank products.

Users can paste reviews, search the pre-seeded SQLite database, chat with the GlowSense Q&A Assistant, or upload custom review CSVs to analyze and compare their own datasets.

---

📸 App Screenshots

---

🚀 Key Features

• 📝 Paste Review Analysis: Paste raw skincare/beauty reviews directly to extract sentiment scores, top skincare topics, skin-type clues (e.g., dry, oily, sensitive), and keyword frequency counts.
• 🔍 Product Lookup: Search by brand, name, or product ID to pull a comprehensive detail sheet: review summaries, YouTube review search links, Amazon/Sephora/Ulta deal finders, ingredients list, and highlights.
• 💬 GlowSense Q&A Assistant: Ask questions about the database and the project (e.g., "How many reviews are in the database?", "Explain how the Glow Score is computed", or "Tell me about the tech stack").
• 📊 Compare Products (Custom CSV uploads): Link and analyze custom customer reviews and product details from uploaded files. Includes:
  • Single CSV (Reviews + details combined).
  • Double CSV (Reviews CSV linked to a Products CSV via a product identifier key).
  • 📥 Save to Database: Bulk-import uploaded files directly into the local SQLite database.
  • Targeted Comparison: Select specific products from a dropdown to run sentiment analysis and render interactive comparison charts on a subset of data (preventing page crashes on large uploads).
• 📥 Seed Data Manager: Seed the database with high-quality reviews from Sephora, Ulta, and Amazon Hugging Face datasets.

---

🧠 Data Science & Sentiment Analysis Logic

1. Lexical Keyword-Based Sentiment Engine

To maintain high explainability and avoid heavy ML dependencies, GlowSense uses a custom lexical tokenization engine:

• Tokenization & Cleaning: Normalizes review text by converting to lowercase, removing punctuation, and filtering out common English stopwords.
• Keyword Matching: Scores text against curated lists of beauty-specific sentiment words:
  • Positive signals (e.g., dewy, hydrating, calm, glowing, smooth, soft).
  • Concern signals (e.g., breakout, acne, burn, irritation, redness, peeling, dry).
• Sentiment Labeling:
  • Mostly positive (score >= 0.25)
  • Mostly negative / concerns found (score <= -0.25)
  • Mixed (between -0.25 and 0.25)

2. Explainable Multi-Factor Glow Score

The Glow Score (0–100) is computed using a dynamically weighted formula that aggregates four dimensions of a product:

1. 🧠 Sentiment Score (40%): Calculated from tokenized customer reviews.
2. ⭐ Product Rating (30%): Average star rating of the product normalized to a 100-point scale.
3. 💰 Price Value (15%): Dynamically compares the product's price against the average price of all products within the same category (e.g., moisturizers). Products cheaper than the category average get a bonus; expensive items scale down.
4. 🏷️ Brand Reputation (15%): Average star rating of all products sold under the same brand in the database.

Note: To ensure statistical reliability, GlowScore scoring only applies to products with 10 or more reviews.

---

📥 Hugging Face & Seeding Pipeline

GlowSense supports seeding reviews from Sephora (CSV), Ulta (CSV), and Amazon (Hugging Face Hub):

• Amazon Dataset: Utilizes the amazon_us_reviews dataset (under the Personal_Care_Appliances_v1_00 or Beauty category).
• Data Streaming: Streams reviews from Hugging Face using the datasets library to avoid downloading massive multi-gigabyte files to local disk.
• 10+ Reviews Filter: Counts product review frequencies dynamically during ingestion. It discards products with fewer than 10 reviews, ensuring only products with significant data are stored in the database.
• Bulk Database Insertion: Writes batches in bulk to the SQLite database via insert_products_bulk() and insert_reviews_bulk() for maximum speed and efficiency.

---

🛠️ Installation & Setup

Prerequisites

• Python 3.8 to 3.12 (Python 3.10+ recommended)
• SQLite3

1. Clone & Navigate

git clone https://github.com/shahidabatool/GlowSense.git
cd GlowSense

2. Install Packages

pip install -r requirements.txt

3. Run the App

streamlit run app.py

The app will launch and open in your default browser at http://localhost:8502.

---

📂 Project Structure

• app.py: Main Streamlit app containing all tabs, page styling, and UI controllers.
• analyzer.py: Contains the sentiment scoring, topic tagging, recommendation generator, and multi-factor Glow Score calculations.
• database.py: Handles SQLite database initialization, table indices, bulk insertion, query helpers, and stats calculation.
• seed_data.py: Script and UI logic to read and import Sephora, Ulta, and Amazon Hugging Face reviews.
• test_analyzer.py: Assert-based test suite verifying tokenization, sentiment calculations, and Glow Score logic.
• requirements.txt: Python package list (streamlit, pandas, altair, datasets, etc.).