Explainable Recommender Systems Evaluation

Adding explanations to recommender systems is said to have multiple benefits, such as increasing user trust or system transparency. However, the evaluation of explanations is still an open research topic with no commonly agreed on guidelines or standards. Comparing different explanation approaches and deciding which approach is the best fit for a given use case is, therfore, challenging.

This project aims to move toward an evaluation approach that allows a researcher to compare the strengths and weaknesses of multiple explanation approaches for recommender systems and investigate their tradeoffs. Showing the same explanation to different users will likely evoke different effects. Even if the same user sees the same explanation, but in different contexts, the effect might vary. Thus, the first step to move into this direction is to understand what factors impact the effects that explanations have on a user.

Abraham Bernstein, Oana Inel, Kathrin Wardatzky