04-800-B   Recommender Systems

Course description

Information overload is a characteristic of our society today. This has made decision-making more complicated than ever before. For example, choosing which restaurant to go to for the day’s meal is further complicated by a variety of dishes on the same menu. In the entertainment industry, one has to choose whether to read a book or watch a movie and, if so, which one. Recommender systems have come in handy as far as overcoming such challenges is concerned. Recommender systems are now largely used, particularly in eCommerce websites, for easing the information search and discovery processes and increasing customer fidelity and conversion rates. They attempt to guide people into making decisions based on their preferences and personalities and by mimicking the choices of people similar to them.

Recommendation systems are part of machine learning or artificial intelligence (AI) algorithms. Students should gain basic knowledge about the field of web recommendation systems and other main recommender systems methods and tools as well as their theoretical foundations. The students will familiarize themselves with the concept of recommender systems and the scientific underpinnings of the latter in different domains.

Students will also gain the necessary tools and techniques to analyze visual and statistical data, build models, and present findings to support data-driven decisions as data science engineers. Students will be involved in hands-on projects such as choosing a topic to work on and getting corresponding data and sources. For example, students can choose from recommender systems in transformer-banking products, healthcare, restaurants, fashion, books, movies, music, and more. For the final project, students will work on RecSys challenges organized by members of the ACM Conference on Recommender Systems. This experience will also give students the skills required for positions in academia and research institutes as well as roles involving network analysis in entrepreneurship and commercial research.

Learning objectives

Starting from basic information retrieval concepts, we will investigate basic techniques for information filtering in recommender systems. We will provide the student with a rich and comprehensive catalog of information search tools that can be exploited in the design and implementation of a specific website such as eCommerce or eGovernment applications for travel, health, or tourism.

Outcomes

Upon completing this course, students will be able to:

• Use the necessary tools and techniques to analyze visual and statistical data, build models, and present findings to support data-driven decisions
• Describe the concepts of Recommender Systems and explain the challenges involved
• Recommend appropriate techniques when faced with a recommendation task, as they will have acquired important data science engineering skills
• Obtain hands-on experience implementing existing methods and evaluating them over real datasets

Content details

Module 1: Introduction and Basic Concept in RS

• Problem domain
• Purpose and success criteria
• Ratings vs. Implicit feedback

Module 2: Paradigms of Recommender Systems

• Collaborative Filtering
• Content-based Filtering
• Advanced Topics in Collaborative Filtering
• Knowledge-Based Recommendations

Module 3: Social-based Recommender Systems

• The basic concept of Social Network Analysis
• Recommender Systems with social regularization
• Trust-aware recommender systems
• Online Consumer Decision Making

Module 4: Group Recommender Systems

• Introduction to group recommender systems
• Recommending sequences
• Modeling satisfaction
• Incorporating group attributes
• Explaining group recommendations
• Evaluating group recommender systems 

Module 5: Other Approaches

• Constraint-based Recommenders
• Hybrid recommendation approaches

Module 6: Hands-on Sessions (Project)

• Designing Real-World Recommender Systems
• Building a Recommender System 

Module 7: Deep Learning for Recommender Systems

• The basic concept of Deep Learning
• The deep learning era of RecSys
• Matrix factorization as embedding learning
• Learning item embeddings
• Word2Vec, Paragraph2vec, doc2vec and Prod2Vec models
• Deep collaborative filtering and CF with Neural Networks
• Autoencoders for recommendation

Prerequisites 

Faculty

Carine Pierrette Mukamakuza