Final Report on Filter Bubbles – Is Facebook Biased?

Facebook Algorithm exposed: the story of Mohammed Abadi

We performed a research on Facebook’s algorithm systems in relation to filter bubbles. The reason behind conducting such an investigation is that most of us are already aware of the existence of filter bubbles but do not know the way they truly work. Most importantly however, we are oblivious to its impact on our daily thoughts and actions. We therefore chose to use data journalism techniques to get a sense of how Facebook algorithms work. Our method was to create a fake profile, feeding it scripted information and collecting the data through the harvesting tool, Fbtrex.

 

The Profile

We created Mohammed Abadi, a twenty-five-year-old aspiring tattoo artist who was born and raised in Baghdad, Iraq. Mohammed obtained an Arts degree at the College of Fine Arts in Baghdad, and afterwards moved to Berlin where he attended the Bard College in order to write his master’s thesis. Currently, he is trying to find a suitable job in Berlin but Mohammed’s main goal is to pursue his dream to become a professional tattoo artist. He is also very close to his family and his culture, interested and involved in Arabic cuisine. He is a practicing Muslim and has a passion for Arabic calligraphy; he therefore aims to take attention away from the sometimes over-provocative and vulgar styles of some tattoo artists. Mohammed is also very interested in cars and their evolvement alongside developing technology, with a special interest in Porsche. Mohammed is a proponent of heteronormativity and does not agree with the LGBT movement and the beliefs they share.

In order to create a realistic persona, we researched places from his childhood city, Baghdad. We liked the Facebook pages of popular cafes, museums and popular news networks related to Iraq. Moreover, in terms of his political and religious interests, we liked and shared posts created by ‘Al Jazeera’, ‘Iraqi News’, ‘Iraq Solidarity News’ and Koranic pages such as Koran al Karim’ which shares daily Quran surahs. Our persona was meant to be a modern, young, artistic man who practices his religion faithfully and who is very politically aware and involved, especially when the subject revolves around Islamophobia and people’s misperception of the Muslim religion. However, we still tried to maintain a more accepted modernity of Mohammed by sharing posts on his creative passions such as beautiful calligraphy and music as we did not want his page to be purely political and religious.

Creating a fake persona on Facebook is ethically questionable, because although we are trying to veer away from stereotypes, the persona we are creating still lives up to a large degree to these stereotypes surrounding race and religion. Another reason why this research is ethically questionable is the project itself. Creating a fake persona is not only misleading for Facebook algorithm systems but for other users as well. The result is a biased feed along with preconceived notions of who a person should be according to their Facebook likes and shares. This can also be problematic when Facebook users believe in the existence of our persona. Lastly, there is always a possibility that real users will emotionally connect to the fake persona, which is another unethical outcome. However, we find that it is for the greater good that Facebook’s filter bubble is being exposed.

 

Extracting the Data

For the extraction of the data we used a tool called Fbtrex, which is a project managed as a GPL free software community and is part of another project: Facebook tracking.exposed. The latter functions as an extension of Google Chrome or Mozilla Firefox. Its aim is to help users track and filter Facebook’s algorithm system. The main goal of this project is to create a fairer online space for both users and developers. Inspired by the peer to peer free internet of the past, tracking.exposed has created its own manifesto in which it describes its main goals and addresses ethical questions relating to the project.

In particular, Fbtrex is a tool aimed to expose Facebook’s potential bias and unfair algorithm systems that limit the information to which users are being exposed on a daily basis, namely, filter bubbles. Among media students like us, this knowledge about Facebook’s recommendation system helps us understand the importance in using such tools to spread awareness and eventually improve these algorithms. Hopefully in the future, this might result in fairer filtering systems and a less secretive online world. As social media users, we are being exposed to only a fraction of what we would see on our feeds if it weren’t for the selection led by these algorithms and filter bubbles. We used Fbtrex to gather and harvest data from our own personal Facebook profiles, then created a fake profile page in order to highlight how different and personalised Facebook feeds become. One of the downsides of Fbtrex is that the tool doesn’t collect all data. The tool exclusively collects timeline posts that are public, cutting out a large chunk of uncollected yet vital data; for example, friend requests. This is unfortunate because friend requests can be very important in the process of getting deeper into a filter bubble.

We inserted the harvested data into an excel worksheet for further analysis. We categorised it and arranged it transparently and easily as possible. When the data was processed, we uploaded it into rawgraphs.io to make this abundance of information more understandable.

Findings

As weeks went by, we saw interesting things happening on Mohammed’s profile. Below we present the findings in numbers:

 

-News                     201          41,1 %

 

-Entertainment          11           2,2 %

 

-Culture                  50           10,2 %

 

-Religion                 37           7,5 %

 

-Food                      1              0,2 %

 

-Cars                      27             5,5%

 

-Tattoo                    30            6,1 %

 

-School                   2               0,4 %

 

 

-Government          4               0,8 %

 

-Brand                    8               1,6 %

 

-Personal               120            24,4 %

 

Total                      491          100 %

 

media bubble chart

The main finding was the uncountable amounts of friend requests he received from Muslim men around the world, especially from Bangladesh and Iraq. Their profiles indicated a rather radical approach to their culture. This showed a bias in Facebook’s recommendation systems, which was something that we wanted to see in our results.

Mohammed never liked radical pages or videos, nor did he watch any radical content but was simply sharing news about the mistreat of Muslims around the world. Facebook however, only recommended him Islamic content and people who they thought he could potentially know because of their shared faith. This data definitely followed the narrative we thought this project would follow once choosing a persona like Mohammed: Facebook’s algorithm tends to put people in the same social circle, group them together depending on their ethnicity or religion, and perhaps a bias towards the Muslim faith. Although Mohammed was not radical, he was suggested to make significantly more radical Facebook friends.

Facebook showed Mohammed a fair amount of posts from Middle Eastern news media channels, comparing to news posts of Western media. This is particularly interesting because we fed the profile equally with Western and Middle Eastern media and still there was significantly a more Middle Eastern news in our feed.

This became more intriguing when we looked into more detail at how we fed the profile with cultural posts. We spent a considerate amount of time liking cultural posts, along with content about cars and tattoo posts. Despite the effort, we did this type of content returning on our timeline or data.

 

bubble graph

Figure 2Posts categorized by type of post and who posted it

Overall, Facebook draws people deeper and deeper in the bubble and focuses on a few factors instead of all the things a profile is interested in. This can be related to the fact that the algorithm is looking for posts which have a lot of traffic, those posts are more probable to be controversial, so the algorithm tries to show as much disputable posts as possible.

What we see in the graphs is that the timeline of Mohammed mostly consisted of single posts. The number of pictures and group posts came on a solid second place.

 

Data Journalism Reflection

In the times of ongoing digital developments, rapidly increasing production of data and fake news, it is crucial to be able to support journalistic stories with data, numbers and statistics. Data journalism is a subfield of traditional journalism, with a main method of working with data sets and information in order to derive stories. It serves to make citizens more aware and involved with the social matters, while using open source data to help make stories more reliable.

Our research fits in the field of data journalism because we tried to make a story out of a collection of raw data in order to expose a problem which would otherwise be kept secret. Additionally, this is a good example of investigating from a user perspective; there was no algorithm study into Facebook; a black, closed box in essence. This research can be conducted by anyone with even basic computer knowledge and an ability to use Excel and visualization tools in order to represent the data. The assignment would fall under the field of data journalism only if it would have been done on a bigger scale and if it would try to expose online patterns or stories unknown to the general public.

From the beginning, this course seemed exciting and interesting. A new way to look at articles, online content and news networks, we acquired a more critical angle towards the way things are written and especially the way data is used to backup information or attract readers. Whilst learning about several different ways of conducting data journalism and constructing our own fake profile, we certainly became more aware of how the world of data journalism works and the importance it carries.
Throughout the classes we obtained several skills that were helpful in conducting our own research. For example, the visualisation and explanation of data through graphs gave us a solid idea of what we wanted to do in our assignment. Moreover, none of us had a clear vision of what data journalism was in the first place, hence learning about these skills and hearing from professionals turned out to be very interesting.
In practice, this helped us in doing our assignment with more ease. At first, we were not completely on the same page as to how we should start our project, disagreeing entirely on who our persona should be. In the end however, everyone focused on achieving the same goal.

Although we commenced this project with the hopes of finding interesting results that would expose Facebook’s bias, we are still satisfied with these findings; the friend requests from somewhat radical men proving our initial point.
Overall, the project and the course itself gave us a new perspective on journalism and data. Most importantly, it gave us new ways of approaching the data that’s available to us and use that to hone our academic skills.

 

Group Reflection 

At the beginning of the course, we divided the tasks in our group according to each team member’s individual skills. We constantly tried to work accordingly to our functions, enabling every group member to efficiently contribute to the project.  Most of the content was created and edited by Lana, the storyteller. Furthermore, her passion for writing and telling stories was reflected in each blogposts. Malik, our finder, was mostly involved with searching for sources concerning the information we were including in our project. Moreover, he was constantly looking for evidence in the form of data, collecting information with Fbtrex and critically examining this data.  The shapers of our group, Marijn and Thomas had an important role within our project, shaping the team work. They focused on the division of tasks and the quality of our work. Visualizer of the group, Lorenzo was focused on visualizing our project and presenting the data in aesthetic, informative and accurate images. He was mostly concerned with the visual and aesthetic aspects of our presentation and blog posts and was responsible for creating the graphs that present our findings. Natalia, who was the team-worker of our group contributed to all functions, while making sure we make the deadlines and upload the assignments in the right place. She played an important role in managing the group, dividing task and creating a good atmosphere within the group. Moreover, she coordinated the schedule of our project and was the person of contact if there were any issues within the project.

Overall, we tried to stick to our roles, so each team member could unfold their individual strengths and skills and feel comfortable with their tasks. Nevertheless, we tried to escape this comfort zone at times involving each member of the group in all aspects of the project.

As we were working together to make our Facebook profile interesting and to find out more about Facebook’s filter bubbles, it was essential to function as a group critically collecting, examining and evaluating the data we gathered.

In conclusion we learned a fair amount about data journalism and how one can with little research materials conduct quite a good research just through the analysis of data.

Posted in Uncategorized