Recommender Systems

The availability of enormous amounts information made possible today by the Internet presents its users and providers with an exciting challenge: how do users find information which is specific and relevant to them and how do providers link topics and user groups appropriately?

Recommender systems provide a solution to these issues and have revolutionized e-commerce, as well as the entertainment and news industries over the past 20 years.

Students know added value, application areas and requirement specifications (e.g. scalability, quality, explainability and representation of recommendations) of recommender systems.

Students understand what type of information is used to personalize recommendations (e.g., "read", "click", "buy", "rate"), where caution is needed in interpreting it, and how sparse information is used.

Students know the conceptual assumptions (e.g. necessity of usage or rating history), advantages and disadvantages (e.g. model-based vs memory-based) as well as mathematical basics of content-based and collaborative filtering (e.g. UBCF, IBCF and SVD) recommenders and are able to implement, compare and hybridize different types of recommender systems.

Students understand the difference of metrics to estimate customer preferences (MAE, NMAE, RMSE), assess the relevance of recommendations (Precision, Recall, F1, MAP), measure their relative ranking (nDCG, MRR) and how their business value is assessed and adjusted based on coverage, diversity and serendipity.