Master theses
Visual evaluation of recommender systems
Author
Jan Šafařík
Year
2022
Type
Master thesis
Supervisor
Ing. Vojtěch Vančura
Reviewers
Rodrigo Augusto da Silva Alves, Ph.D.
Department
Summary
The thesis aims to explore new ways of evaluating recommender systems, mainly with the help of visualization tools. In the current research, when evaluating the success of a model, the emphasis is mainly on the accuracy of recommendations but less on other important metrics, such as the diversity of recommended items, the extent of catalog coverage, or the opportunity of discovering novel items at the expense of bestsellers. Moreover, evaluation results are typically presented only in the form of aggregated values, and their distribution through the validation set of users is not further investigated, which is essential for revealing the unfairness of the model and discovering various biases. Therefore, the thesis offers a set of new approaches that combine visualization techniques with existing offline evaluation methods and are implemented as part of a framework made available under an open-source license to other researchers.