Accurate, efficient, and explainable modelling of context-dependent preferences using matrix factorisation

Nowadays, consumers are overwhelmed with a large number of products/services available on the web, and recommender systems help them choose the right products/services to buy. They also help businesses match their products/services to the right consumers and hence increase their profits. A dilemma faced in preference modelling in recommender systems is whether to choose an understandable/explainable simple system while sacrificing accuracy, or an accurate complex model while sacrificing explainability. This thesis addresses the problem of making understandable, highly accurate, and efficient recommendations to the users.