How dangdailoi/recommendation-system works
Project type and purpose
• This is a full-stack web application (Flask) that provides product recommendations and search capabilities for an e-commerce site.
• It implements deep learning (CNN‐based content filtering) and reinforcement learning (Deep Q-Learning) for personalization, plus text and image search (Whoosh, FAISS).
File structure
• Top-level:
• app.py – Flask entrypoint/routes
• db/ – data persistence (SQLAlchemy ORM, raw SQL schema)
• recommendation_system/ – ML-based recommender modules (dcbf.py, dqn.py)
• search_engine/ – search modules (text_search.py, image_search.py, Whoosh schema)
• static/ & templates/ – UI assets (CSS, JS, HTML)
• model.json – configuration for model paths or parameters
• Patterns:
• MVC-style division: templates+static = View, app.py = Controller, db & ML modules = Model & Services
• ML components separated into their own packages
• Config files (requirements.txt, model.json) and scripts (db/dbo.sql)
Key insights from README
• Application flow: user → Flask app → (search_engine OR recommendation_system) → database & ML model files
• Dependencies: Flask, SQLAlchemy, TensorFlow, FAISS, Whoosh
• External assets: large pre-trained models stored externally (.h5, .npy)
Main components
a. Frontend (Browser):
• HTML templates + static CSS/JS
b. Web Server (Flask App):
• app.py (routes, REST endpoints)
• model.json (config loader)
c. Data Layer:
• Relational DB (SQLAlchemy ORM via db/dbo.py)
• Raw SQL scripts (db/dbo.sql)
d. Search Engine Services:
• Text search (Whoosh via search_engine/text_search.py + schema.py)
• Image search (FAISS or custom via search_engine/image_search.py)
e. Recommendation Services:
• Deep content-based filtering (recommendation_system/dcbf.py)
• Deep Q-Learning RL recommender (recommendation_system/dqn.py)
• Pre-trained model storage (.h5, .npy, user‐provided)
f. External & Auxiliary:
• Model storage on file system or object storage
• Optional data crawl/ingestion component
Relationships & interactions
• User → (HTTP) → Flask App
• Flask App routes
• “/search” → invokes search_engine modules → queries Whoosh index or FAISS index
• “/recommend” → invokes recommendation_system modules → loads models → computes suggestions
• Database reads/writes for user profiles, product metadata, logs
• Flask App ↔ Database via SQLAlchemy
• Search Engines ↔ Search Index (Whoosh files or FAISS vectors on disk)
• Recommenders ↔ Model files on disk
Architectural patterns & principles
• Layered Architecture: Presentation → Application → Services (Search, Recommendation) → Data
• Separation of concerns: distinct packages for search vs. recommendation vs. persistence
• Dependency injection/config via model.json
• Stateless web server; stateful data & index storage
Detail
• Both text and image search flows.
• How user interactions (search vs. recommendation) trigger different services.
• Where configuration (model.json) is read.