Fundamentals of Recommender Systems

Lesson 1 • Business Value and Use Cases

Examines how recommendations drive engagement, revenue, and retention. Connects system design choices to measurable business outcomes.

Lesson 2 • Taxonomy of Recommender Approaches

Surveys collaborative, content-based, and hybrid paradigms at a high level. Provides a mental map for organising all techniques covered later.

Lesson 3 • The Recommendation Problem Defined

Defines the core task of matching users to items under uncertainty. Establishes vocabulary used throughout the course.

Lesson 4 • Data Landscape for Recommendations

Identifies the types of data that feed recommender systems. Prepares students to assess data availability in real projects.

Lesson 1 • Data Cleaning and Normalisation

Removes noise, duplicates, and outliers that distort model training. Introduces normalisation techniques that improve rating comparability across users.

Lesson 2 • Building the User-Item Matrix

Transforms raw logs into a structured interaction matrix. Addresses sparsity, which is the defining challenge of collaborative filtering data.

Lesson 3 • Collecting User Interaction Data

Covers instrumentation strategies for capturing clicks, ratings, and dwell time. Connects data quality at collection time to downstream model accuracy.

Lesson 4 • Feature Engineering for Items and Users

Creates derived features from metadata and behaviour for content-based and hybrid models. Bridges raw attributes to model-consumable representations.

Lesson 1 • Similarity Metrics for Neighbourhoods

Introduces cosine, Pearson, and Jaccard similarity for user and item neighbourhoods. Builds intuition for choosing metrics based on data type.

Lesson 2 • Matrix Factorisation Fundamentals

Decomposes the interaction matrix into latent user and item factors. Introduces SVD and its practical approximations for sparse data.

Lesson 3 • Handling Cold Start in Collaborative Filtering

Addresses the new-user and new-item problem within collaborative paradigms. Introduces fallback strategies and hybrid bridges.

Lesson 4 • User-Based Collaborative Filtering

Generates predictions by aggregating ratings from similar users. Covers neighbour selection, weighting, and prediction formulas.

Lesson 5 • Item-Based Collaborative Filtering

Shifts similarity computation to the item space for greater stability. Explains why item-based methods often outperform user-based at scale.

Lesson 1 • Similarity-Based Recommendation

Ranks candidate items by similarity to the user profile vector. Connects similarity metrics from Chapter 3 to the content-based context.

Lesson 2 • Strengths and Limitations of Content Filtering

Critically evaluates where content-based methods excel and where they fail. Motivates the hybrid approaches introduced in the next chapter.

Lesson 3 • Item Representation and Feature Spaces

Converts item attributes into vector representations suitable for similarity computation. Establishes the feature space that drives content-based matching.

Lesson 4 • Text-Based Recommendation with NLP

Applies natural language processing to extract semantic features from item descriptions. Enables recommendations based on textual content beyond keywords.

Lesson 1 • Hybrid Design Patterns

Catalogues switching, weighted, mixed, and cascade hybrid strategies. Provides a decision framework for selecting the right pattern per use case.

Lesson 2 • Context-Aware Recommendation

Incorporates time, location, and device context into hybrid models. Shows how contextual signals shift recommendation relevance.

Lesson 3 • Ensemble Methods for Ranking

Applies bagging, boosting, and stacking to combine ranker outputs. Demonstrates accuracy gains over single-model baselines.

Lesson 4 • Factorisation Machines

Introduces factorisation machines as a unified model for sparse feature interactions. Covers the FM equation and its extension to higher-order interactions.

Lesson 1 • Limitations of Offline Evaluation

Explains why offline metrics often fail to predict online performance. Bridges to online evaluation methods covered in later chapters.

Lesson 2 • Beyond Accuracy: Diversity and Novelty

Measures list diversity, novelty, and serendipity as complements to accuracy. Motivates multi-objective evaluation in production systems.

Lesson 3 • Ranking and Top-N Metrics

Introduces precision, recall, NDCG, and MAP for evaluating ranked lists. Connects metric choice to the business goal of surfacing relevant items.

Lesson 4 • Accuracy Metrics for Rating Prediction

Covers RMSE, MAE, and their variants for evaluating predicted ratings. Explains when each metric is appropriate and how to interpret differences.

Lesson 1 • Autoencoders for Collaborative Filtering

Uses encoder-decoder networks to reconstruct sparse rating vectors. Covers denoising autoencoders as a regularisation strategy.

Lesson 2 • Sequence and Session-Based Models

Models user behaviour as a sequence to capture short-term intent. Introduces RNNs, GRUs, and attention mechanisms for session recommendation.

Lesson 3 • Graph Neural Networks for Recommendations

Represents users and items as nodes in a bipartite graph for message passing. Captures higher-order collaborative signals beyond direct interactions.

Lesson 4 • Neural Collaborative Filtering

Replaces the dot-product interaction with a multilayer perceptron for richer user-item modelling. Demonstrates accuracy gains over classical matrix factorisation.

Lesson 1 • A/B Testing and Online Experimentation

Designs controlled online experiments to measure recommendation impact. Covers traffic splitting, guardrail metrics, and experiment duration.

Lesson 2 • Two-Stage Retrieval and Ranking

Separates candidate generation from fine-grained ranking for scalability. Explains how each stage optimises different objectives under latency constraints.

Lesson 3 • Model Monitoring and Drift Detection

Tracks model performance and data distribution shifts in production. Establishes alerting and retraining triggers to maintain recommendation quality.

Lesson 4 • Scalability and Infrastructure Patterns

Addresses horizontal scaling, batch vs. real-time pipelines, and cost optimisation. Prepares students to make infrastructure decisions for large-scale systems.