CP4285: Modern Recommendation Systems

Schedule

Mon, 01 Jan 0001 00:00:00 +0000

Class Meetings

Day	Time	Venue
Every Tuesday	10:00-12:00 SGT

Weekly Schedule

Week	Date / Period	Focus	Assessment Due
1	Tue, 11 Aug 2026	Recommendation Problems and Classical Methods
2	Tue, 18 Aug 2026	Latent Factor Models
3	Tue, 25 Aug 2026	Evaluation of Recommendation Systems
4	Tue, 1 Sep 2026	Neural Recommendation Models
5	Tue, 8 Sep 2026	Sequential and Session-Based Recommendation
6	Tue, 15 Sep 2026	Retrieval and Ranking Architectures
Recess Week	Tue, 22 Sep 2026	No class
7	Tue, 29 Sep 2026	Project Design Critique Workshop	Project design critique
8	Tue, 6 Oct 2026	Learning-to-Rank
9	Tue, 13 Oct 2026	Graph-Based Recommendation
10	Tue, 20 Oct 2026	Multi-Objective Recommendation
11	Tue, 27 Oct 2026	Exploration and Online Learning
12	Tue, 3 Nov 2026	LLMs, Generative Recommendation, and Research Frontiers
13	Tue, 10 Nov 2026	Final Project Presentations	Final project presentation or report (depending on STePS participation)
Reading Week	Sat, 14 Nov 2026-Fri, 20 Nov 2026	Reading week
Examination Week 1	Mon, 23 Nov 2026	Examination period	Final exam, 13:00-15:00 SGT, venue to be announced
Examination Week 2	Sat, 28 Nov 2026-Sat, 5 Dec 2026	Examination period

Weekly Details

Week 1: Recommendation Problems and Classical Methods (11-17 Aug 2026)

Topics:

Recommendation tasks
Explicit vs implicit feedback
Popularity-based recommendation
Content-based recommendation
Collaborative filtering

Ethics thread: popularity bias, exposure inequality, and platform incentives.

Learning outcomes:

Formulate recommendation problems.
Differentiate recommendation paradigms.
Analyze cold-start challenges.
Explain how recommendation objectives affect exposure.

Week 2: Latent Factor Models (18-24 Aug 2026)

Topics:

Matrix factorization
User-item embeddings
Bias models
Ranking objectives, including Bayesian Personalized Ranking

Ethics thread: bias encoded in historical interactions, representation, and interpretability.

Learning outcomes:

Explain latent-factor models.
Train embedding-based recommenders.
Compare prediction and ranking objectives.
Discuss risks of learning from historical behavior.

Week 3: Evaluation of Recommendation Systems (25-31 Aug 2026)

Topics:

Offline evaluation
Precision@K
Recall@K
MAP
NDCG
Business metrics

Ethics thread: metrics as value choices, and accuracy versus user welfare.

Learning outcomes:

Design evaluation protocols.
Compute ranking metrics.
Critique metric selection.
Explain limitations of offline evaluation.

Week 4: Neural Recommendation Models (1-7 Sep 2026)

Topics:

Neural collaborative filtering
Deep ranking models
Representation learning

Ethics thread: explainability versus performance, and transparency concerns.

Learning outcomes:

Build neural recommenders.
Compare neural and latent-factor approaches.
Analyze explainability challenges.
Assess trade-offs between complexity and transparency.

Week 5: Sequential and Session-Based Recommendation (8-14 Sep 2026)

Topics:

User sequences
GRU4Rec
SASRec
Transformer recommenders

Ethics thread: engagement optimization and behavioral manipulation risks.

Learning outcomes:

Model temporal preferences.
Implement next-item prediction.
Compare static and dynamic representations.
Critically assess engagement-driven objectives.

Week 6: Retrieval and Ranking Architectures (15-21 Sep 2026)

Topics:

Candidate generation
Retrieval
Ranking
Re-ranking
Two-tower architectures

Ethics thread: visibility allocation and stakeholder impacts.

Learning outcomes:

Explain industrial recommendation pipelines.
Design retrieval-ranking architectures.
Analyze scalability trade-offs.
Evaluate how ranking affects different stakeholders.

Week 7: Project Design Critique Workshop (29 Sep-5 Oct 2026)

Student deliverables:

Application domain
Dataset
User problem
Recommendation objective
Baselines
Evaluation plan
Ethical risks

Ethics thread: feasibility, bias, evaluation design, and project scope.

Peer critique themes:

Is the recommendation task well-defined?
Are the evaluation metrics appropriate?
What biases may emerge?
Is the problem feasible?

Learning outcomes:

Defend recommendation-system designs.
Critique evaluation strategies.
Identify ethical risks early.
Refine project scope based on feedback.

Suggested weight: 5-10% participation or milestone grade.

Week 8: Learning-to-Rank (6-12 Oct 2026)

Topics:

Pointwise ranking
Pairwise ranking
Listwise ranking
LambdaRank intuition

Ethics thread: position bias and fair ranking.

Learning outcomes:

Formulate ranking objectives.
Compare ranking approaches.
Analyze position bias.
Discuss fairness implications of ranking.

Week 9: Graph-Based Recommendation (13-19 Oct 2026)

Topics:

User-item graphs
Graph embeddings
Graph neural networks
LightGCN

Ethics thread: homophily, echo chambers, and community amplification.

Learning outcomes:

Represent recommendations as graph problems.
Explain graph propagation.
Build graph-based recommenders.
Analyze risks of graph-driven feedback loops.

Week 10: Multi-Objective Recommendation (20-26 Oct 2026)

Topics:

Diversity
Novelty
Serendipity
Coverage
Long-term satisfaction

Ethics thread: balancing stakeholder interests.

Learning outcomes:

Define non-accuracy objectives.
Measure diversity and novelty.
Design multi-objective recommenders.
Justify objective trade-offs.

Week 11: Exploration and Online Learning (27 Oct-2 Nov 2026)

Topics:

Multi-armed bandits
Contextual bandits
Exploration-exploitation
Feedback loops

Ethics thread: online experimentation and fair exposure.

Learning outcomes:

Explain exploration strategies.
Design adaptive recommendation policies.
Analyze recommendation feedback loops.
Discuss ethical implications of experimentation.

Week 12: LLMs, Generative Recommendation, and Research Frontiers (3-9 Nov 2026)

Topics:

Conversational recommendation
LLM-enhanced recommendation
Retrieval-augmented recommendation
Foundation models
Causal recommendation
Future directions

Ethics thread: trust, hallucination, persuasive AI, and governance.

Learning outcomes:

Explain modern recommendation research directions.
Evaluate LLM-based recommendation systems.
Critique emerging approaches.
Identify open research challenges.

Week 13: Final Project Presentations (10 Nov 2026)

Required presentation components:

Problem formulation
Dataset and preprocessing
Baselines
Proposed recommender
Evaluation results
Error analysis
Ethical analysis
Future improvements

Ethics thread: technical, product, and societal considerations.

Learning outcomes:

Present recommendation-system designs professionally.
Defend technical decisions.
Interpret evaluation results critically.
Integrate technical, product, and societal considerations.

Assignments

Mon, 01 Jan 0001 00:00:00 +0000

Coursework is organized around steady participation, written critique, project work, and short checks on core concepts. The final exam assesses individual understanding across the semester.

Coursework Components

Component	Weightage	Description
Class Participation	10%	Active contribution to seminars, project critique, peer feedback, and in-class discussion.
Essays	20%	Short written analyses of recommender-system methods, evaluation choices, and social implications.
Project / Group Project	30%	Team project covering problem formulation, dataset preparation, baselines, model design, evaluation, error analysis, ethical analysis, and final presentation.
Quizzes / Tests	10%	Short checks on foundational concepts, metrics, and model design choices.
Final Exam	30%	Individual final examination covering the full course.

Submission Notes

Submission channels, file formats, and deadlines will be announced by Min.
Group submissions should clearly state each member’s contribution.
Written work should include enough methodological detail for Min to assess assumptions, implementation choices, evaluation design, and ethical reasoning.

Group Project

A key part of mastering any skill is practicing it beyond the formal algorithmic basis. Projects form an integral part of the assessment (30% of total marks). Student groups should have 5 to 6 members and will be partially self-assembled (into initial subgroups of size 1 to 3) and then assembled by Min into final groups, with project preferences and expertise taken into account. Min will correspond with groups and check in on a regular basis. It is your responsibility to ensure that your group meets with Min, not Min’s responsibility to chase you.

Min will propose a set of suitable recommendation system datasets for student groups to work with. Details will be released in Canvas Files.

Note that performance on macroscopic metrics alone is not the critical factor in your grade. We primarily evaluate with respect to the interesting and well-motivated ideas your team employs to solve the task, the quality of your evaluation, and your ethical analysis.

Project Milestones

Milestone	Week	Description	Weight
Project Mini-team Declaration	Week 4–5	Declare your initial subgroup (1–3 students), skills, and project interests via the Canvas survey.	—
Project Design Critique	Week 7	In-class workshop. Each team presents their application domain, dataset, user problem, recommendation objective, baselines, evaluation plan, and ethical risks for peer critique.	Part of Participation
Final Project Presentation	Week 13	Teams present their implemented recommender, evaluation results, error analysis, ethical analysis, and future improvements.	Part of Project (30%)
Final Project Report	Exam Period	Written report covering problem formulation, dataset preparation, baselines, model design, evaluation, error analysis, and ethical analysis.	Part of Project (30%)

Project Topics and Datasets

Min will release a curated list of recommendation system datasets suitable for the project. You are expected to select one dataset and define a clear recommendation problem around it. Your project should include:

A well-defined user problem and recommendation objective.
At least one classical baseline (e.g., collaborative filtering, matrix factorization).
At least one neural or advanced model.
A rigorous offline evaluation using appropriate metrics (e.g., NDCG, Recall@K, MRR).
An error analysis identifying failure modes.
An ethical analysis addressing bias, fairness, privacy, or stakeholder impact.

Compute Resources

You may use the SoC Compute Cluster for your project work. Details on how to access the cluster will be provided in Canvas Announcements. You may also use Google Colab or other cloud platforms, but ensure your experiments are reproducible.

Peer Assessment via TEAMMATES

This course uses for intra-team peer assessment to evaluate individual contribution and calibrate project grades. There are two rounds:

Formative Assessment (Week 7, ungraded): A mid-project peer review to give your team early feedback on contribution and collaboration. Results are for your team’s reflection only and do not affect your grade.
Summative Assessment (Week 13, graded): A final peer review conducted at the end of the course. Results will be used to moderate individual project grades — outstanding contributors may receive a grade uplift, while those who have not contributed adequately may receive a reduced project grade.

You will receive an invitation to TEAMMATES via your NUS email. Please complete both assessments by their respective deadlines.

Academic Honesty for Projects

Group projects are collaborative by nature. However, each member is expected to contribute meaningfully. Please refer to the for the full academic honesty policy, including the No-Sponge Rule. AI tools may be used for the project but must be documented appropriately.

Frequently Asked Questions (FAQ)

Mon, 01 Jan 0001 00:00:00 +0000

For questions about lecture material, please use the relevant weekly discussion thread.

Administrivia

What are the prerequisites for CP4285? The formal prerequisite is completion of at least 120 units. (Introduction to AI and Machine Learning) or equivalent is also required. Students who have not taken CS2109S but have equivalent background in neural machine learning may seek approval from Min.
What is the format of this course? Seminars are mandatory, physical face-to-face sessions. The course is taught by Kan Min-Yen. The class meets every Tuesday, 10:00–12:00, at Seminar Room 12, COM3-01-21. There are no tutorial sessions for this course.
Will lectures be recorded? Lectures will be recorded where technology permits. Recordings are made available to facilitate revision only and are not a substitute for attendance. The standard NUS expectation is that all activities are face-to-face.
I’m doing an ATAP/SIP/FYP related to recommendation systems. Am I allowed to take the course? Generally yes. However, if the course is oversubscribed, you will need to make an official appeal. Please contact Min directly if you have concerns about your eligibility.
I’m on exchange. Can I take this course? Exchange students are welcome. Please ensure you meet the prerequisites and follow the standard NUS exchange enrolment process.
Can I audit the course? Auditing is not officially supported. If you are interested, please contact Min.

Grading and Assessments

How is the course graded? Please refer to the for the full breakdown: Final Exam (30%), Group Project (30%), Essays (20%), Quizzes/Tests (10%), Class Participation (10%).
Are essays individual or group work? Essays are individual take-home assignments (each worth 10%). AI tools are permitted as a resource, but no collaboration with other students is allowed. You must submit an AI declaration with each essay; where requested by Min, you must provide full documentation of your AI use. Please refer to the for the full AI use policy.
What is the group project about? Please refer to the for full details. In brief, your team will select a recommendation system dataset, implement classical and neural baselines, evaluate them rigorously, and conduct an ethical analysis.
How are project groups formed? You will first self-assemble into initial subgroups of 1–3 students via the Project Mini-team Declaration survey (due Week 5). Min will then assemble final groups of 5–6 students, taking project preferences and expertise into account.
Will my participation grade be visible on Canvas? No. Your participation grade will not be available to you on Canvas until final grades are released.
What happens if I miss the pre-flight or midterm survey? These surveys contribute to your participation marks. Missing them will result in a loss of those marks. Please complete them by their due dates.

Exam

When is the final exam? The final exam is on Mon, 23 Nov 2026, 13:00–15:00 SGT. It is worth 30% of your total marks.
What is the exam format? The exam will be conducted on the secure Examplify platform. Only Windows and MacOS laptops are supported (no tablets or iPads). The exam will consist mostly of MCQ, MRQ, and short answer questions. No AI tools are permitted during the exam.
What if I don’t have a compatible laptop? You may request a loaner laptop from CTLT via the Loaner Laptop Quiz, which opens in Week 11. Note that the number of available laptops is limited.

Course Content and Prerequisites

What are the prerequisites for CP4285? The formal prerequisite is completion of at least 120 units. (Introduction to AI and Machine Learning) or equivalent is also required. Students who have not taken CS2109S but have equivalent background in neural machine learning may seek approval from Min.
What programming language will we use? We will use Python 3.11 or newer, with libraries such as PyTorch, Scikit-learn, and RecBole for recommendation system tasks.
Will we cover ethical issues in this course? Yes. Ethical considerations — including bias, fairness, privacy, exposure, transparency, and stakeholder impact — are interwoven throughout the curriculum as a core thread, not an afterthought.

Grading

Mon, 01 Jan 0001 00:00:00 +0000

Assessment Breakdown

S/N	CA Component	Weightage	Notes
1	Class Participation	10%	Participation in seminars, project critique, peer feedback, and class discussion.
2	Essays	20%	Written analysis and critique assignments.
3	Project / Group Project	30%	Group project with design critique, implementation, evaluation, final presentation, and report components.
4	Quizzes / Tests	10%	Short checks on course concepts and methods.
10	Final Exam	30%	Final examination on 23 Nov 2026, 13:00–15:00 SGT. Venue to be announced.

Total: 100%.

Attendance at seminars is mandatory and participation is part of your overall mark. The participation grade will be marked as a combination of your in-seminar participation, your project critique contributions, peer feedback, and your participation in class surveys (e.g., pre-flight and midterm surveys).

Academic Honesty Policy

Please note that we enforce these policies vigorously. While we hate wasting time with these problems, we have to be fair to everyone in the class, and as such, you are advised to pay attention to these rules and follow them strictly.

Collaboration is a very good thing. Students are encouraged to work together and to teach each other on appropriate course components (n.b., not for individual essays). On the other hand, cheating is considered a very serious offence. Please don’t do it! Concern about cheating creates an unpleasant environment for everyone. You will be automatically reported to the Vice-Dean of Academic Affairs if you are caught; no exceptions will be made for any infractions, no matter how slight the offence.

So how do you draw the line between collaboration and cheating? Here is a reasonable set of ground-rules. Failure to understand and follow these rules will constitute cheating, and will be dealt with as per University guidelines. We will be enforcing the policy vigorously and strictly.

You should already be familiar with the University’s (Office of Student Conduct, NUS). If you haven’t yet, read it now.

The Pokémon Go Rule: You are free to meet with fellow student(s) or AI agents and discuss assignments with them. Writing on a board or shared piece of paper is acceptable during the meeting; however, you may not take any written (electronic or otherwise) record away from the meeting. This applies when the assignment is supposed to be an individual effort (i.e., individual essays in CP4285). After the meeting, engage in a half-hour of mind-numbing activity before starting to work on the assignment. This will assure that you are able to reconstruct what you learned from the meeting by yourself, using your own brain.

The Freedom of Information Rule: To assure that all collaboration is on the level, you must always write the name(s) of your collaborators on your submission. Failure to adequately acknowledge your contributors is at best a lapse of professional etiquette, and at worst it is plagiarism. Plagiarism is a form of cheating.

The No-Sponge Rule: In intra-team collaboration where the group, as a whole, produces a single product, each member of the team must actively contribute. Members of the group have the responsibility (1) to not tolerate anyone who is putting forth no effort (being a sponge) and (2) to not let anyone who is making a good faith effort fall through a crack (to help weaker team members come up to speed so they can contribute). We want to know about dysfunctional group situations as early as possible. To encourage everyone to participate fully, we make sure that every student is given an opportunity to explain and justify their group’s approach.

The Pokémon Go Rule and Freedom of Information Rule are adapted from Surendar Chandra’s course at the University of Georgia, who in turn acknowledges Prof. Carla Ellis and Prof. Amin Vahdat at Duke University for the original policy formulation. The origin of the rule, known as the Gilligan’s Island Rule, can be traced to Prof. Dymond at York University (1984). The No-Sponge Rule is adapted from the same tradition.

AI Use Policy

This course follows the (NUS Centre for Teaching, Learning and Technology, 2024). The course-specific application of that policy is as follows:

Component	AI Permitted?	Conditions
Group Project	Yes	AI use is permitted and encouraged where appropriate. You must document all AI use in your submission. Share your AI usage approach with your group and document it collectively.
In-seminar participation	Yes	AI use is permitted. You must be able to explain and justify any AI-assisted contributions in your own words.
Individual Essays (take-home, individual)	Yes (with declaration)	AI tools are permitted as a resource. However, no collaboration with other students is allowed. You must submit an AI declaration with each essay. Where requested by Min, you must provide full documentation of your AI use (e.g., prompts, outputs, and how they were incorporated). The essay remains an assessment of your own critical thinking; AI may assist but must not substitute your analysis.
Final Examination	No	The examination is conducted on Examplify in a closed, supervised environment. No AI tools are permitted.

For the Group Project, if you do not document your AI use, it implies that you did not use AI; any doubts we have may be investigated and prosecuted. For Individual Essays, an AI declaration is mandatory regardless of whether AI was used; failure to submit a declaration will be treated as a policy violation. Your submission of work and your e-signature during submission acknowledges your compliance with and understanding of these rules.

Where in doubt, please raise your concerns with Min before attempting any potentially non-compliant actions. Unfortunately, we have had to prosecute and fail students who have not obeyed these rules.

Late Submissions

We do not accept late submissions for essays or project milestones unless you have a valid reason (e.g., medical certificate). Please contact Min as early as possible if you anticipate difficulties meeting a deadline. Extensions will be granted at Min’s discretion.