In the ever-expansive world of various choices, such as the multiple different movies or books, recommendation systems offer users an easy way to navigate through their options. Additionally, the expansion of social media has led more people to form groups for specific activities, and consecutively, group recommender systems have emerged as a popular research area.
“In contrast to recommendations for single users, group recommendations involve a much greater degree of complexity since the systems are responsible for balancing the often-conflicting interests of all group members”, Maria Stratigi says.
Considering the impact on user perception and the sensitivity of tasks like e-health recommendations, it's important to carefully assess the recommendation generation process. As a result, it has become increasingly necessary to develop recommendations that adhere to various responsibility constraints such as fairness, which corresponds to a lack of bias, and transparency which facilitates an understanding of the system.
“If these constraints are followed, group recommendation systems become more complex. It is even more challenging if they are to consider a sequence of recommendations rather than each recommendation as a separate process”, Stratigi clarifies.
Sequential group recommendation systems add a new layer of complexity
Intuitively, the group recommendation system should consider the historical interactions between itself and the group, and adjust its recommendations in accordance with the impact of previous suggestions. This observation leads to the emergence of a new type of recommender, called sequential group recommendation systems.
“A sequential group recommender should record all recommendations provided to the group, along with group members' satisfaction, and consider both when producing suggestions for the next round. Multiple recommendation rounds add a new layer of complexity for achieving fair group recommendations”, Stratigi tells.
Supposing each group member experiences a degree of satisfaction with a provided group recommendation list, the system should provide recommendations without bias against any group member throughout a series of recommendation rounds. In other words, the same group member should not always be the least satisfied.
In the dissertation, Stratigi also examines transparency in recommendations, commonly referred to as explanations in recommendations. Frequently, recommendation systems provide explanations in conjunction with recommendations so that the user can then gain an understanding of the rationale behind a suggestion.
Users can be presented with explanations of recommendations automatically or after providing feedback. In the latter case, the feedback often takes the form of a query from the user to the system.
“For example, a question can be either why an item is suggested or why an item was not suggested. Many works explore the first case. However, questions concerning the non-existence of a data item remain unexplored. These questions are called why-not questions and this dissertation details their definition as well as presents a model for answering them”, Stratigi says.
Public defence on Monday 11 September
The doctoral dissertation of MSc Maria Stratigi in the field of computer science, titled Group Recommendations with Responsibility Constraints, will be publicly examined at the Faculty of Information Technology and Communication Sciences at Tampere University at 12:00 pm on Monday 11.9.2023 at Tampere University City centre campus, Pinni B, auditorium 1100 (Kalevantie 4, 33100 Tampere, Finland). The Opponent will be Professor Tanca Letizia, Politecnico di Milano. The Custos will be Associate Professor Konstantinos Stefanidis, Faculty of Information Technology and Communication Sciences.
The doctoral dissertation is available online.
The event can be followed via remote connection.
Photograph: Alexakis Konstantinos