If you wish to pursue your thesis in collaboration with our group, please contact konstantinos.stefanidis@tuni.fi with a tentative research proposal. Feel free to use the following suggestions as a source of inspiration.
Topics in Recommender Systems
Create new algorithms for generating responsible recommendations for individual users and groups of users, i.e., recommendations that ensure fairness, diversity and transparency.
- Group Recommendations: Generalised recommendations to suggest data items to groups of users
- Sequential Recommendations: Keep past interactions to ensure fair results for the group in a sequence of suggestions
- Fairness in Recommender Systems & Rankings: Use Variational Autoencoders to increase the fairness (mitigate the position bias) in sequential recommendations.
- Diversity in Recommender Systems: Ensure diversity & coverage with respect to both data items and user characteristics in single-user and group recommendations.
- Why-not Questions in Recommender Systems — Transparency: Formalise the problem of why-not questions in recommender systems.
- Ratings vs. Textual reviews in Recommenders: Investigate the effectiveness of using the sentiment of users textual reviews as the users ratings.
- Equality-enhancing Technology for the Flipped Classroom – Alternatives to traditional lecture-based instruction using active learning and flipped classroom approaches, emphasizing on self-study of teaching material and peer collaboration.
- Fair group formation techniques: Allocate the best members in a fair way between the teams when considering multidisciplinary requirements
Topics in Entity Resolution
Entity Resolution (ER), especially, when decisions about which entity descriptions should be resolved first, need to be made with respect to a given budget (e.g., a computational cost or a number of data passes)
- Formalise progressive ER as an optimization problem in which we are not simply interested in maximizing the number of resolved entities for a given budget, but the diversity of the entity graph resulting after merging the matching descriptions.
- Fast approximate solutions to our diversity-driven ER problem with formal guarantees.
- Ensure fairness: The ratio of matching entities with respect to the available ground truth remains equal for all subgroups.
- Enhance transparency: Provide explanations.
- Why two descriptions were considered candidate matches.
- Why two descriptions were found to match.
- Why were the merged nodes preferred to be returned over alternative plans respecting the same budget.