Project: Neural Collaborative Filtering based Recommendation System

From CLIENTOCLARIFY.AI PVT LTD (Jun 2019 - Dec 2021) [cite: 235, 238]

Business Problem Statement

The challenge was to build a highly accurate and real-time personalized recommendation system for movies, TV shows, and web series, especially for local broadcast channels and Video-on-Demand (VOD) services, with efficient data handling for large datasets. [cite: 239, 240, 241, 243]

Solution

• Collected 1.76 PB of data and built data pipelines and dashboards for real-time data visualization and recent trends in movies, TV shows, and web series. [cite: 239]
• Built a Softmax DNN recommendation system model for movies, web series, and TV shows with 90% accuracy for local broadcast channels. [cite: 240]
• Built the Neural Collaborative Filtering (NCF) Model with 98% accuracy on live data, incorporating different tuning optimization algorithms and parameters. [cite: 241]
• Built Matrix Factorization, Large-scale Retrieval, and multiple candidate generators using different sources, followed by a re-ranking approach to filter candidates. [cite: 242]
• Achieved 10% faster real-time personalized recommendations user experiences at scale in VOD. [cite: 243]

Tools & Frameworks Used

• Softmax DNN [cite: 240]
• Neural Collaborative Filtering (NCF) [cite: 241]
• Matrix Factorization [cite: 242]
• Data Pipelines [cite: 239]
• Dashboards (for data visualization) [cite: 239, 356]
• Python [cite: 338]
• NumPy, Scikit-learn, TensorFlow, PyTorch, Keras (general AI/ML libraries) [cite: 295, 351]