Online services have the capacity of learning the preferences and interests of individual customers based on their online activity. Using this knowledge, the online services can be personalized. This personalization filter is referred to as the filter bubble, and it is built from the actions of the user by algorithms run by the services. However, the algorithms used by the online services are not public and carefully kept private, whilst the filter bubble of users strongly influences the information they access, which has a big impact in society. Therefore, properly understanding the algorithms for filter bubbles is an important and open question.
The goal of MyBubble is modeling the influence of algorithms in the users’ filter bubble in the online advertising ecosystem. To this end, a methodology developed by researchers of the MyBubble team will be leveraged. This methodology allows creating “personas”, “bots” that visit carefully selected websites, referred to as “training websites” that assign them a specific behavior. Once the persona profile has been created, the researchers will make the persona visit “training” and “control” websites. “Control websites” are those from where the ads shown to the created persona will be collected. The persona will keep visiting the training websites so that it retains its behavior during the whole execution of the experiment.
The experiments to be run with this methodology will allow the unveiling of existing algorithmic biases, how the personalization modifies the behavior of the online advertising algorithms under different personas, and how the algorithms change the filter bubble when the persona changes its behavior.