JultikaUniversity of Oulu repository

Abstract

The term filter bubble refers to a phenomenon in which a recommendation system fails to offer diverse or novel content, and instead offers content that reinforces particular belief systems. Filter bubbles are considered harmful when they restrict users’ exposure to diverse content and thereby reinforce potentially harmful or misinformed ideologies, contribute to the spread of misinformation, and foster the creation of echo chambers. This thesis proposes a solution to counteract the effects of filter bubbles by providing users with the option to switch content feeds with their least similar users’ feed.

The solution was achieved by substituting the correlation coefficient used in collaborative filtering recommendation systems. An application was developed to simulate post recommendations for users, initially employing a traditional collaborative filtering system. This was then followed by a collaborative filtering system that recommended posts based on the likes of the least similar user to the current user. User engagement metrics and cognitive mapping metrics were used to evaluate this solution. If the solution did not negatively affect user engagement and demonstrated an ability to increase the diversity of users’ bias perception and promote a more nuanced understanding of bias within the social media application, it met the requirements of these metrics.

There was an overall increase in user engagement after the users’ feed was swapped. Moreover, the users’ perception of bias became more diversified, indicating that the solution prompted a broader awareness of bias within the social media application users were engaging with. Based on these results, the proposed solution was deemed as potentially effective in addressing the filter bubble problem. The solution’s viability was established solely within a simulated environment. To determine its real-world applicability, it requires further testing in a naturalistic environment with more participants.