AI for Personalized Recommendations Full Syllabus

Module 1: Introduction to Recommendation Systems

• What is personalization? Why does it matter?
• Real-world examples (Amazon, Netflix, Spotify, Coursera)
• Types of recommendation systems
  • Content-based
  • Collaborative filtering
  • Hybrid approaches
• Evaluation metrics overview (Precision, Recall, NDCG, MAP, etc.)

Module 2: Data Collection & Preprocessing

• Collecting data: explicit vs implicit feedback
• Cleaning user-item interaction datasets
• Dealing with sparsity and cold start problem
• Feature engineering for users and items
• Data splitting strategies for RS (train/test/time-based)

Module 3: Content-Based Filtering

• Building item profiles using metadata (genre, tags, description)
• Building user profiles from interaction history
• Cosine similarity, Euclidean distance, and TF-IDF
• Use of NLP in item representation (titles, descriptions, etc.)

Module 4: Collaborative Filtering

• User-User collaborative filtering
• Item-Item collaborative filtering
• Matrix Factorization
  • SVD (Singular Value Decomposition)
  • NMF (Non-negative Matrix Factorization)
• Limitations and advantages of CF

Module 5: Deep Learning for Recommendations

• Neural Collaborative Filtering (NCF)
• Embedding layers for users and items
• Autoencoders for dimensionality reduction
• DeepFM (Factorization Machines + DNNs)
• Use of CNNs/RNNs in sequential recommendations

Module 6: Sequence & Session-based Recommendations

• Time-aware & session-based recommendation needs
• Recurrent Neural Networks (RNN, GRU, LSTM)
• Transformer models (SASRec, BERT4Rec)
• Next-item prediction logic

Module 7: Evaluation of Recommendation Systems

• Offline evaluation metrics
  • Precision@K, Recall@K
  • MAP, NDCG, MRR
• Online evaluation (A/B testing, click-through-rate)
• Diversity, Novelty, Serendipity, and Coverage

Module 8: Hybrid Recommendation Systems

• Combining content-based and collaborative filtering
• Weighted and switching hybrids
• Model-based vs heuristic-based hybrid RS
• Case studies of Netflix & Amazon hybrids

Module 9: Context-Aware Recommendation Systems

• Incorporating location, time, device, weather, etc.
• Contextual bandits
• Multi-armed bandit approaches
• Real-world application examples: Uber, Swiggy, YouTube

Module 10: Graph-Based Recommendation Systems

• User-item graphs and bipartite networks
• Node embeddings using DeepWalk, Node2Vec
• Graph Neural Networks (GNNs) for recommendation
• Knowledge Graph-enhanced recommendations

Module 11: Tools, Frameworks & Deployment

• Libraries:
  • Surprise
  • LightFM
  • TensorFlow Recommenders
  • PyTorch + PyTorch Lightning
• APIs: TMDB, IMDB, Spotify, Goodreads
• Deploying RS with Flask, FastAPI, Streamlit
• Cloud deployment (AWS/GCP/Azure)

Module 12: Advanced Techniques

• Reinforcement Learning-based recommendations
• Meta-learning for cold-start problems
• Fairness, Bias, and Explainability in RS
• Federated learning for privacy-preserving personalization