PV254 Recommender Systems

About the course

Recommender systems are very active area of both research and application (well known applications include Amazon, Netflix). This course covers basic principles of recommender systems, particularly with focus on collaborative filtering (recommendations based on people behaviour) and on educational applications (including discussion of projects developed at our faculty). The focus is also on practical experience (a project).

Schedule

Preliminary schedule:

Lectures (with discussions):
- Feb 20: Introduction, overview
- Feb 27: Collaborative filtering, forming teams for projects
- Mar 5: Other recommendation techniques, discussion of project ideas
- Mar 12: Evaluation, discussion of project evaluation
- Mar 19: Personalization in education, case studies, practical experiences
- Mar 26: Practical Aspects, Case Studies
Work on projects, individual consultations: April
Final project presentations: May 7, May 14

Projects

There are two basic options - an "applied" project and a "research" project. (More options will be discussed during the lecture.)

Development of a simple recommender system

Project for teams of 1-4 students.
Goal: To build a simple recommender system.
The focus should be on functionality (not on user interface). The system should include enough content and functionality to be "interesting".
Suggestions (will be discussed in more detail during lectures):

"short text" recommendations: jokes, quotes, poetry, baby names, recipes, ...
"local" recommenations, travel: restaurants, cultural events, places in Brno, holiday locations, countries to visit, tourist attractions, geocaching, ...
educational recommenations: courses (MU, MOOC), foreign language vocabulary, learning materials, ...
product recommendation (specialized for a particular domain): board games, books for children, wines, beers, specific movie genre, ...
personalized guides: TV program, museum guide, ...

Model development and evalution

Individual project
Goal: For a given dataset develop a model for predicting user ratings / student performance. Evaluate the model (compare to previous models). Provide visualizations of the domain (similarities between "items").
Specific data sets are provided, together with some basic guidelines for model development and evaluation. Two types of data are available:

Data about movies.
Educational data (geography practice).

Materials

More datasets:

Recommendation and Ratings Public Data Sets For Machine Learning
Datasets referenced in an overview "Characterization of Public Datasets for Recommender Systems" (Çano, Erion and Morisio, Maurizio, 2015).

Sources

Main recommended sources:

Recommender Systems: An Introduction. D. Jannach, M. Zanker, A. Felfernig, G. Friedrich, 2010.
- slides
Recommender Systems Handbook. F. Ricci, L. Rokach, B. Shapira, P. B. Kantor, 2015 (second edition).
Note: Electronic version available at MU
Introduction to Recommender Systems (video lectures from Coursera course)
Introduction to Recommender Systems: A 4-hour lecture, Xavier Amatriain, Machine Learning Summer School 2014 @ CMU
Recommender Systems: The Textbook. Charu C. Aggarwal, 2016.
Note: Electronic version available at MU