How fuongfotfet/movie-recommendation-system works
Project Type and Purpose
• This is a full-fledged movie recommendation platform that integrates multiple recommendation algorithms:
– Neural Collaborative Filtering (NCF)
– KNN + FAISS (Approximate Nearest Neighbors)
– Content-based Filtering
• Besides the recommendation algorithms and data processing, a web-based prototype is provided for user interactions, allowing users to select genres, pick movies, and receive real-time recommendations.
Analyzing the File Structure
• The project is structured to separate different recommendation approaches into their own directories:
– “Neural Collaborative Filtering” contains notebooks, scripts, and saved model checkpoints.
– “Collaborative Filtering – KNN and ANN” includes a FAISS-based implementation as well as KNN scripts and accompanying Jupyter notebooks.
– “Content-Based Filtering” contains the content-based implementation.
• The “web” directory is dedicated to the front-end and back-end components:
– The “web/main/frontend” directory houses the HTML, CSS, and JavaScript files that form the user interface prototype.
• Data used by different modules is located within “data” directories across several modules – these include CSV files with user data, movie metadata, ratings, etc.
• Other accompanying directories and files (e.g., “document” with a Report PDF) provide complementary documentation and reports.
A. Data Sources:
– Files in various “data” folders (CSV files like u_data.csv, u_genre.csv, train/test splits for NCF, etc.)
– Movie metadata (possibly extended via TMDB integration)
B. Recommendation Engines:
– Neural Collaborative Filtering: Modules (NCF.ipynb, NCF.py, & trained models) representing deep learning-based recommendations.
– KNN + FAISS: Modules (faiss_final.py, knn.py, and their corresponding notebooks) handling high-dimensional vector similarity searches.
– Content-based Filtering: Scripts (content_based.ipynb and content_based.py) analyzing movie metadata for similarity.
C. Web Application Layer:
– Backend: Flask/Python app (app.py) tying in recommendation logic.
• Contains submodules for the different recommendation approaches within “web/main/backend”.
• Additional backend scripts for obtaining movie posters, trending items, top genres or ratings.
– Frontend: User interfaces in “web/main/frontend” that allow users to select genres, choose movies, and view recommendations.
D. External Integrations and Dependencies:
– FAISS library for conducting fast approximate nearest-neighbor searches.
– TensorFlow (for Neural Collaborative Filtering).
– Python libraries such as numpy, pandas, and scikit-learn.
– Optional integration with TMDB for enhanced movie metadata.
Elements
• Clear labels for each component, for example:
– “Data Source – CSV files”
– “Neural Collaborative Filtering Engine”
– “KNN + FAISS Engine”
– “Content-Based Filtering Engine”
– “Flask Application (API & Routing)”
– “Frontend: Select Genres, Movie Selection, Recommendation Display”
– Inputs to processing engines, outputs from engines routed back to API layers, and then to the UI.
• Annotations or notes to indicate dependencies and technology stacks (e.g., “FAISS” next to the KNN module, “TensorFlow” with NCF).
Happy diagramming!