Investigating the Effect of Recommender System Algorithms on the Amplification of Disinformation

Abstract

Recommender systems are used by many services that everyone encounters in their daily life. These include platforms such as Google, Netflix, TikTok etc. These platforms use recommender systems to recommend the items to users that it expects the user will prefer. They often allow its users to add their own items to the input of its recommender system e.g. by posting a video on TikTok. This makes these platforms vulnerable to disinformation. Disinformation is information that is deliberately created to mislead its target audience [3]. A person with malicious intent could use those recommender systems to spread disinformation. This can be done by adding an item that was created with the intention to mislead the users of the recommender system. This thesis will investigate the amplification of disinformation in recommender systems. This project will focus on several issues that might influence the amount of disinformation that is recommended. Such as a comparison of various recommender system algorithms, changes in the popularity of fake items and changes in the characteristics of the input data. This will be done by creating a simulation of the spread of disinformation. In this project fake items will be created and added to an original dataset. These fake items will represent disinformation. This is then used as the input for several recommender systems. The resulting recommendations will then be used to determine if there were fake items recommended and if so how many. This could give some insight into how disinformation and its characteristics affect the recommendations that are produced by recommender systems.