10.00 – 10.45: Trial lecture. Title: “Generative AI for Recommendation Systems”
12.15 – 15.15: Public defence
The dissertation is available in ODA Open digital Archive (oda.oslomet.no).
The trial lecture and public defence will also be streamed live:
- Webinar ID: 659 9655 0528
- Passcode: 0125
Ordinary opponents:
- First opponent: Associate Professor Amira Mouakher, ESPACE-DEV laboratory, Université de Perpignan Via Domitia, France
- Second opponent: Professor Sadok Ben Yahia, Center of Industrial Software, University of Southern Denmark, Denmark
- Leader of the evaluation committee: Associate Professor Kazi Shah Nawaz Ripon, Department of Computer Science, OsloMet, Norway
Leader of the public defense: Professor André Brodtkorb, Head of Department of Computer Science OsloMet
Supervisors:
- Main supervisor: Professor Anis Yazidi, Department of Computer Science, OsloMet
- Co-supervisor: Associate Professor Youcef Djenouri, Department of Computer Science, University of South-Eastern Norway
- Co-supervisor: Professor Paal Engelstad, Department of Computer Science, OsloMet
Abstract
Group recommendation systems (GRS) are designed to find what a group of people likes and suggest things they will enjoy together, such as choosing a movie, deciding where to dine, or planning a trip. These systems aim to find options that suit everyone's preferences, making group activities more enjoyable and harmonious. However, combining the diverse tastes of group members into a single recommendation is a challenging task.
This research focuses on improving GRS by developing new methods to make recommendations both fair and precise. It explores how groups make decisions and uses advanced algorithms to predict what people with varying preferences will enjoy together. To tackle these challenges, the study introduces innovative ways to address key issues, such as dealing with limited preference data, grouping users based on shared interests, and ensuring that everyone’s voice is fairly represented in the final recommendation.
Key contributions of this work include:
- A novel method for predicting missing preference data, improving recommendations even when little information is available.
- Advanced techniques for identifying similarities between users using tools like graph neural networks and matrix factorization, which outperform traditional methods.
- New clustering methods for grouping users with similar tastes, ensuring fairer recommendations.
- A consensus-reaching approach inspired by real-life dynamics, such as personality traits, to reflect how groups make decisions.
- Aggregation methods that balance individual preferences to create a cohesive group choice, using concepts like the Shapley value and Wonderful Life Utility.
By combining these approaches, this research makes significant strides toward creating smarter, fairer, and more accurate group recommendation systems. These advancements pave the way for better tools to enhance social experiences, making group decisions easier and more enjoyable.
