Recommender Systems

Full course description

Recommender systems play an important role in helping to mediate many of our everyday decisions and choices, including the music we listen to, the news that we read, and even the people that we date. They do this by learning from our past interactions, inferring our interests and documenting our preferences. To make the right suggestions at the right time recommender systems must not only understand our preferences but also our current needs and perhaps our immediate intent. Thus, the core focus of most recommender systems is devoted to profiling users and matching items based on these profiles and current context.

Much of the research to date on recommender systems has focussed on the engineering and evaluation of core recommendation algorithms. Researchers have developed a variety of approaches to harness different forms of preference data in the pursuit of more accurate recommendations. For example, researchers have used simple ratings for collaborative, rich meta-data for content-based methods, and even the opinions and sentiment expressed within user-generated reviews.

When evaluating recommender systems, there has been a heavy emphasis on measuring the accuracy of suggestions, or the error of predictions. However, in practice it is important to consider evaluation metrics beyond accuracy, such as diversity, novelty, and serendipity. This in turn has led to increased attention being given to the nature of the interactions between users and recommender systems, and the influence that the user interface and interaction style can have on user behaviour and the overall recommendation experience. This course focuses on:

• Non-personalized and Stereotype-based Recommender Systems
• Classical recommender systems algorithms, e.g., Content-based Filtering, Collaborative-based Filtering
• Offline Evaluation e.g., protocols, criteria, metrics
• User-centered evaluation
• Interfaces and interaction in Recommender systems, e.g., explanations and conversational recommender systems
• Ethics, bias, and fairness in recommender systems
• Advanced methods, e.g., Matrix Factorization, Hybrid recommender systems, Contextual Recommender systems

This is an optional course: Third year students choose three electives per period out of the optional courses during period 1 and 2.