Counterfactual Recommendation for the Evaluation of Recommender Systems

The evaluation of recommender systems nowadays is mostly based on so-called offline evaluations. Based on a historic dataset of user interactions with the system (e.g., a user having rated a set of movies), we simulate user behavior and evaluate the performance of a given recommender algorithm based on this simulation. Generally, these datasets are collected in a system with an already active recommender system, so the resulting data is biased in this sense.With counterfactual learning methods, one can address the question of how well a new recommender algorithm would have performed if it had been used instead of the original algorithm. In this thesis, we aim to investigate the use of counterfactual learning methods for the evaluation of recommender systems.