Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/5D31F8F5-A062-401A-B97F-6F4920D33C97 | ||
| Year | 2024 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Panagiotis-Angelos Ganotis, "Item Fairness in network friendly recommendations systems", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2024 https://doi.org/10.26233/heallink.tuc.99212 | ||
| Appears in Collections |
Recommendation Systems (RS) play a crucial role in shaping narratives and influencing people’s choices. This technological field has seen a lot of research in recent years, both due to its utility to users and also due to its ethical implications. Most content and social media platforms use recommendation systems to give users the most relevant and engaging options based on their past preferences. Recent work suggests integrating network performance into the design of recommendation system algorithms, demonstrating considerable potential. However, network-friendly adaptations of RS algorithms may create unfairness for users and content providers. A Network-friendly Recommendation System also introduces bias toward (a smaller pool of) low-cost content, raising fairness concerns for both consumers and creators. Fairness is now considered acomplementary optimization dimension to achieve a ’win-win-win’ situation for network/content providers, users, and content creators. This thesis focuses on exploring the creation of ’content bubbles’ as a specific form of unfairness, which has not been extensively studied in the problem before. At a high level, a ’content bubble’ implies that the selection of items suggested to a user (or group of users) is less diversified in order to facilitate network cost reduction. The main contribution of this thesis is to define appropriate metrics to measure ’content bubble’ and answer the following questions: i) Do NF-RS algorithms foster content bubbles, and ii) Do current fairness metrics effectively address this issue? Our results have indicated that the existing metrics do not fully capture the ’content bubble’ problem and moreover, the tolerance of unfairness in order to establish high network gains may cause recommendations to lack diversity, strengthening the ’content bubble’ effect.
Copyright © DIAS 2013
