CP4285: Modern Recommendation Systems examines the algorithms, data, and design trade-offs behind contemporary recommender systems. The course connects classical recommendation methods with modern neural approaches, and emphasizes practical implementation, careful evaluation, and responsible deployment in real-world settings.
CP4285 covers classical methods, neural architectures, ranking and retrieval pipelines, sequential models, graph-based recommendation, online learning, and emerging LLM-based approaches. Ethical issues are interwoven throughout, including bias, fairness, privacy, exposure, transparency, and stakeholder impact. Students apply these ideas in a hands-on group project requiring problem formulation, dataset work, baselines, model design, evaluation, critique, and final presentation.
The course is hosted by WING.NUS, the Web IR / NLP research group at NUS led by Min.
Prerequisites: CS2109S (Introduction to AI and Machine Learning) or equivalent, and completion of at least 120 units. Students without CS2109S but with equivalent background may seek approval from Min.
Workload: (2-0-0-3-5) — 2 hours lecture, 3 hours projects and assignments, 5 hours preparatory and other work; approximately 10 hours per week.
By the end of the course, students should be able to:
| CLO | Outcome |
|---|---|
| CLO 1 | Explain and compare classical recommendation methods, including collaborative filtering, content-based filtering, and hybrid approaches. |
| CLO 2 | Implement matrix factorization and neural recommendation models using modern deep learning frameworks. |
| CLO 3 | Design and execute rigorous offline evaluation protocols using appropriate ranking metrics. |
| CLO 4 | Analyse the cold-start problem and propose strategies to address it. |
| CLO 5 | Critique recommender systems from fairness, privacy, transparency, and stakeholder impact perspectives. |
| CLO 6 | Explain advanced recommendation architectures including sequential, graph-based, multi-objective, and LLM-enhanced systems. |
| CLO 7 | Design and justify a complete recommendation system pipeline from problem formulation through evaluation and ethical analysis. |
| CLO 8 | Communicate and defend recommendation-system designs and results to a technical audience. |