With TV shows like MTV’s Catfish, creating fake accounts on Facebook has become a recurring topic on the news. It used to be quite easy to create a fake account online, but platforms such as Facebook have recently made multiple changes to its terms and regulations after multiple scandals. The Cambridge Analytica scandal, for example, is one of the most well-known scandals to have inspired these changes. It would be logical to assume that it would be much harder to create a fake account after these changes, which is why we decided to dive into this matter by attempting to develop a filter bubble by playing into Facebook’s algorithm.
The main reason for conducting this experiment was looking at the isolating nature of filter bubbles within Facebook. Eli Pariser, a well-known media scholar and internet activist, who coined the term filter bubbles in his book The Filter Bubble: What The Internet Is Hiding From You,gave a lot of insight regarding the functioning of personalized news feeds. According to Pariser, filter bubbles determine what you see online. Users see a completely personalized assortment of posts and advertisements recommended by the social media platform’s algorithm to persuade the user into engaging with them. Often us users are not aware of inhabiting this filter bubble in which we are alone.
We created Alexander Ivanov, a nationalist Russian expat who recently moved to the Netherlands, to see how the filter bubbles would operate within such a profile and whether it varies much from the filter bubbles we as Facebook users encounter. Unfortunately, Alexander Ivanov did not work out, so a new account with the name Thomas Ivanov was made later on.
Thomas Ivanov is a persona who is not anywhere compatible with ourselves. We chose this method so that there would be a clear distinction between the filter bubbles we inhabit as opposed to his. So naturally, we expected to see very different content in comparison to the content that appears on our own Facebook profiles. However, considering the time available for this experiment we did not expect for much content to appear on his Facebook account. We expected for the content that we received to be very different from our own and the information we acquired to mainly come from Facebook groups or communities. Instead of an extensive research, this is rather a small-scale experiment considering the short time frame we had to operate in.
Bringing the Bot to Life
To start off this experiment we created a Facebook bot. This required setting up a detailed profile for a fictional Russian Nationalist, named Thomas Ivanov. We created him to be a 35-year-old Russian expat who just moved to the Netherlands and to be an active user on Facebook. He likes and shares posts about Russia, in particular military posts, and focuses primarily on events in the Netherlands, which he would join to integrate into the culture and community. Further details regarding his background included Thomas Ivanov being an alumnus at Orenburg State University in Orenburg. This city is known to be nationalistic, which would strengthen his profile of being a Russian Nationalist. Military posts are interesting to look into since a large number of Russian men are enlisted and are heavily influenced by the politics that they form in their early years there. They are assumed to have an inclination towards nationalism, since they fight for their country.
Designing this bot and focusing on specific traits we hoped would eventually manipulate Facebook’s algorithm into placing Thomas Ivanov in a filter bubble of Russian nationalists. The decision was made to investigate the filter bubble of Russian nationalism since it is very different from our personal profiles. One of our group members is able to speak Russian which allowed us to join native Russian groups. Expats are usually very isolated groups, so they tend to flock together and form their own ideas and politics. This allows nationalism to flourish which is why we made this a part of the fake profile.
The Failed First Attempt
Thomas Ivanov was actually not the first attempt of conducting this experiment. Alexander Ivanov was the original bot, but failed to be verified by Facebook. His name, e-mail and Dutch phone number, that we had to provide were accepted straight away. However, his account was rapidly disabled while his profile photo was still in the process of being verified.
This issue was the first obstacle of the experiment and prevented us to post or befriend other people on the website. Factors, such as using a copyrighted picture and an already-used phone number linked to another account were considered, but in hindsight we assumed that the fault might have been in relation to the name. Alexander Ivanov seems to be a very common name, however, by only changing the first name of our second bot it quickly got accepted. According to the Terms of Service of Facebook, authenticity is highly valued and this could be why generic names might not be accepted by the algorithm and their regulations. Yet we can only speculate about the cause of our account being disabled.
Another factor that was different whilst creating the new profile, was accessing the site from a different IP address. We believed that initially creating the profile from the University’s IP address could lead to issues. However, this is not in line with the fact that the first bot, which got deleted, was made from a personal IP address, whereas the second bot was made from the University’s IP address. This contradicted our first thought.
What Stood Out?
Once Thomas Ivanov was created successfully, the obvious absence of advertisements in his timeline caught our immediate attention. Usually while looking at a Facebook profile, advertisements and sponsored posts are shown regularly; some might not cohere to the profile, but they are clearly present. Generally, we find that we perceive advertisements after every two or three posts. Thus, it is notable and interesting that our bot did not receive any advertisements at all. We think this may be due to the account being so new. Perhaps Facebook did not yet have enough data collected in order to personalize advertisements for Thomas Ivanov. However, advertisements could still have been based on his age, sex and/or location.
During the last week of our experiment our bot suddenly started to get recognized by other users. A lot of friend requests were coming in, which changed our complete timeline. Instead of just seeing posts from Facebook groups, personal posts, newly added profile pictures and shared videos from Thomas Ivanov’s friends started showing up. Besides that, the timeline also started to show ads of (mainly Dutch) events. What was strange is that these events didn’t seem to follow Thomas Ivanov’s interests. They were focused on a Dutch audience, while Thomas Ivanov didn’t see any Dutch language posts in his timeline.
Visualising the Data
In order to conduct the experiment, we tracked the bot’s activity on Facebook in two periods of five days. The first period is the stage in which the filter bubble was being developed by the website, and the second period is the stage in which it would become evident that Ivanov entered the Russian Nationalist filter bubble.
Initially, we were planning on comparing the bot with the account of a member from our group, Nadia, who is the most active on Facebook and has a clear-cut political inclination online. The plan was to compare the bot with Nadia’s account to understand if different filter bubbles can compare. However, this plan had to be disregarded as 90% of her data came up as unidentified when we collected it. Since there was almost no data being categorized in either a post, photo, event or group, this could not be used in our experiment. We attempted to collect the data from the other group members as well, to see if that would give us the desired results. We quickly came to the conclusion that we could not use our own personal accounts for this project, as we all encountered the same issue. Due to this problem, we chose to change our research question to how do filter bubbles develop.
After the data from the first five days was collected from Thomas Ivanov’s fake profile, we extracted it with fbtrex, which is a tool to track personalization algorithms in order for people to have control over their online Facebook activity. The data was then converted into a csv.file and subsequently opened in Excel. This spreadsheet provided a coherent overview of all the data acquired over the first period and identified according to their types. These types included posts, photos, events and groups. We then counted all these types, by sorting them in categories on Excel, and entered these in Raw Graphs, which is a tool to take data from spreadsheets and transform them into a visualization. Our initial thoughts were to choose either a scatterplot, boxplot or sunburst, since these turned out to be the easiest to understand and the most popular, according to Kennedy Elliott. Eventually, we chose a circle packing graph as this type of graph is often used to show comparison between values in terms of its proportions. Once we decided on the graph we played around with the colors and dimension for a little bit to determine which suited our approach best.
We then continued to conduct the same experiment with the data collected from the second stage, which showed significantly different results.
In the first graph, we can clearly see that the groups were most evident on the bot’s timeline. Our explanation for this is that filter bubbles are created and are influenced by one’s community. Facebook assumes that whatever your friends like, you like. The fact that a lot of groups were coming up on Thomas Ivanov’s timeline, is an indicator of Facebook trying to gage in what community the bot might be in, by presenting him with a lot of groups. In January 2019, an algorithm change was passed to show users more content from friends and family as opposed to business posts. This is also a possible explanation as to why we saw so many group posts.
There were very little data types other than group posts that were visible on the first Excel spreadsheet. However, the second graph showed much more variety in terms of the data types. Since, the bot was active on Facebook for a longer time, the algorithm had more time to develop a filter bubble for him, which confirms our hypothesis. Thomas Ivanov got exposed to more people in the first stage, in order for him to get acquainted with his community and thus eventually be placed in a filter bubble. In the second stage, more targeted posts in the second stage, which indicates that he was placed in the filter bubble.
Summing It Up
Developing two fake profiles on Facebook proved to be harder nowadays than it used to be. Facebook has become stricter with its verification policy, which is evident from our first failed attempt. Moreover, we noticed that it takes time for advertisements and personalized posts to show up on the timeline of a new profile, which proves our hypothesis of filter bubbles being influenced by the people you interact with. This explains that advertisements are more personalized and specific for profiles that have existed for longer as opposed to a new profile. With Facebook’s new algorithmic update, focusing on more posts by friends and communities, the filter bubbles will once again be influenced rather by your community than news outlets.
We thought that making the bot was not causing anyone much harm, but the idea of creating a fake profile proves how easy it is to pass as a real person – which feels wrong. Taking terms like ‘catfishing’ into consideration, we unanimously agreed that this experiment was questionable considering some ethical standards were not met. Using someone else’s photo and attaching it to a different identity without their consent is outright wrong. It also concerning to us how incredibly easy it still is to create a fake profile.
Focusing on the fbtrex plug-in, we found it not to be the most useful tool for this experiment. There was no control group, making it difficult to obtain any substantial data. We are also skeptical about obtaining useful results with the tool, when it comes to comparing it to older profiles, like the ones of our group members. Especially considering the data from older profiles, given to us by fbtrex, was often not completely identified.
We are also apprehensive about future purposes of this tool. It is not fit for this kind of experiment due to the fact that the tool is suited for quantitative data research, which was not the case for this small group project. Looking into filter bubbles through a bot focused on primarily quantitative data, such as the differences in our newsfeed or general activities.
In our personal opinion this research could be improved through a quantitative approach and perspective. Multiple factors were not controlled, i.e. our lack of a control group, and did not provide reliable data for analysis and comparison between accounts. In addition to this, this experiment was acting as a confirmation bias towards filter bubbles, rather than trying to disprove the hypothesis.
Written by Daniella Bischoff, Femke van Bruinessen, Jasmina Rehman, Nadia Heemskerk, Tara de Gelder and Tessel van Oirsouw. They collectively worked on this journalistic article, dividing tasks where they seemed fit and organised manners in a way that motivated productive teamwork.
Blogpost 3: Filter Bubbles, a Facebook Bot and Lots of Data
After our first fake Facebook account got deleted, we collectively decided to create a new one. Our new bot followed the same directive as the initial one, with as biggest difference that it now actually got to live a life on Facebook. We changed the name from Alexander Ivanov to Thomas Ivanov and again focused his interests on Russian nationalism. We built up a profile with a different picture than before, but obtained it in a similar manner from Shutterstock. At the beginning, we thought that the picture was the cause of our first bot being deleted, but in hindsight we assume it was the name. Even though we only changed the first name, Alexander Ivanov seems to be a very common name which could be a possible cause. Yet we can only speculate about the cause of our account being disabled. Another difference whilst creating the new profile was accessing the site from a different IP address. We believed that initially creating the profile from the University’s IP address would lead to issues. However, the first bot, which got deleted, was made from a personal IP address, whereas the second bot was made from the University’s IP address.
What caught our attention is an obvious absence of advertisements in the timeline of the bot. All of us in the group get advertisements regularly; some are incorrect regarding our interests, but they are clearly present. Generally, we find that we perceive advertisements after every two or three posts. Thus it is notable and interesting that our bot did not receive any advertisements at all. We think this may be due to the account being so new. Perhaps Facebook did not yet have a good enough idea of who Thomas Ivanov is, in order to personalise advertisements for him. However, there should still be advertisements that could be connected to his age, sex and/or location.
Now let’s move on to the data we acquired from our fake profile. Firstly, we extracted the data as a csv.file from the fbtrex tool. Once we had that data in an Excel file, we counted the amount of posts, photos, events and groups. Afterwards, we put all this data in raw graphs; the value corresponding to the amount present of that group in the file. To present our data in an organized way we debated about what graph to use. Our primary thoughts were to chose either a scatterplot, boxplot or sunburst, since these turned out to be the easiest to understand and the most popular, according to Kennedy Elliott. Eventually, we chose a circle packing graph which shows circles representing hierarchies and representing values. This graph shows how certain elements are proportionate to each other. Once we decided on the graph we played around with the colors and dimension for a little bit to determine which suited our approach best.
We were planning on comparing the bot account with the account of Nadia, the most active Facebook user of our group, but when we wanted to put her data in a graph it turned out that a lot of the data wasn’t identified as either post, photo, event or group. This was true for more than 90% of all the data gathered. We then downloaded the data of other members of our group and we encountered the same problem. Three of our group members’ data displayed this issue, whereas with the bot everything was identified even though there was less data. We wonder if this has anything to do with the Facebook accounts being older. This seems to be the only logical explanation for this, since the account of the bot was used on the laptop of a group member whose private data was not identified. It thus could not have been a flaw in the download of the tool or the browser.
Lastly we wanted to share our thoughts of the fbtrex tool we used for this research. We are still a little confused regarding how useful the outcome of our research is. First of all, we want to address that there is no control group in this experiment, making it in our opinion difficult to really obtain any substantial data. We also are skeptical about obtaining useful results of the tool when it comes to comparing it to older profiles (like the one’s of our group members), especially considering the data from older profiles, given to us by fbtrex, was often not completely identified. We hope to find out soon how this tool could be beneficial when figuring out filter bubbles.
Blogpost 2: The Deleted Russian Bot and Other Facebook Experiences
Our plan for our Facebook bot was to set up a profile for Alexander Ivanov, a Russian expat who moved to The Netherlands and is currently learning Dutch. He is a 35-year-old programmer who was actively involved in Facebook, thus liking and sharing posts about Russia and its army in particular. Our intention was for him to be ‘interested’ in events within The Netherlands to get to know the country better. Regarding his background, we came up with the idea that he went to Orenburg State University in Orenburg, which is more nationalistic than other parts of Russia and would explain where his expressions come from. This bot would therefore collectively be designed with the intention of researching the filter bubble of Russian nationalism.
Despite our carefully thought out profile, we did not succeed in creating it. Our bot was disabled by Facebook whilst it was trying to verify the picture, so we came as far as merely submitting general information, such as his name, e-mail and phone number. Because of this, we were unable to even post or follow other people like we intended to. We were consistent with using the same IP address whilst using the bot, so the fault did not lie there. We also considered that the phone number we needed to use to verify the account was perhaps already in use, but after checking it, this claimed not to be true and that the number was not linked to a current Facebook account at the time of making the bot. As a group, we discussed the cause of the unsuccessful Facebook page even further and came to the conclusion that the fault lied in the picture, as it was retrieved from Shutterstock.
If we had succeeded in actually making the Facebook account, we would have expected to see a completely different timeline altered to the specific interests of our bot. Especially considering how different his life and thoughts are to ourselves and also considering the difference in gender.
To us, our research proved that in this present time, it is more difficult to set up a Facebook account than it was before. It seems like their verification process is significantly stricter after issues surrounding previous events such as for example the 2016 presidential elections of the United States of America.
Ethically, we thought that making the bot was not wrong per se, but the idea of making a fake profile proves otherwise. For research purposes, we understand that this was a task that we needed to complete. However, taking terms like ‘catfishing’ into consideration, we unanimously agreed that this research was not ethically justifiable despite using the photo from Shutterstock. Using someone else’s photo and attaching it to a different identity without their consent is outright wrong.
Whilst discussing the fbtrex tool in class we had our apprehension regarding the feature. Now that we have had the feature for a week, we do not see this as a fit tool for this kind of experiment due to the fact that the tool is suited for quantitative data research, which was not the case for this small group-work research. Our research into these filter bubbles was aimed at looking at the overall ‘experience’, meaning we focused primarily on qualitative data, such as the differences in our newsfeed or general activities.
However, our group members have noted that early in January Facebook had changed it’s algorithm to favour content posted by friends rather that any news outlets etc. Upon discussion and comparison of our news feeds we have come to the conclusion that our feeds still primarily focus on friend-activity, events around us, advertisement primarily focused on shopping, as well as random entertainment videos. One of our team members, Nadia, who is most active on Facebook, as well as with a clear-cut political inclination, had more news articles show up than the rest of us. The other team members did not really notice a significant change that stood out. We expected that the people who are more active on Facebook would have a more tailored and different newsfeed than the people who are less active, but we did not find this to be true.
In our personal opinion on how this research could be improved for further studies, rather than just acting as a confirmation bias towards filter bubbles which have been taught to us in lectures, this kind of tool and research must be approached from a quantitative perspective. Multiple factors were not controlled and thus do not provide reliable data for analysis and comparison between accounts.
Blogpost 1: Fake News and False Flags
Our group takes the position that the article ‘Fake News and False Flags by The Bureau of Investigative Journalism does not feature enough explicit quantitative data, in order to be qualified as a ‘data’ journalism project, rather it is primarily based on citizen journalism, namely the single interview of Martin Wells.
The article investigates how the Pentagon – the US Department of Defense – paid British PR firm Bell Pottinger half a billion US dollars to work in Iraq. The purpose was for them to make fake terror and TV segments that were manipulated in the way that it looked like it came from Arabic news sources.
The article makes allusions or hints towards official documents, data or statistics being ‘unearthed’, however, it is not transparent enough: “A document unearthed by the Bureau shows the company was employing almost 300 British and Iraqi staff at one point.” This does not provide the insight into which documents presented this number. Further on, the journalists write: “The Bureau has identified transactions worth $540 million between the Pentagon and Bell Pottinger for information operations and psychological operations on a series of contracts issued from May 2007 to December 2011.” In these numerical statements the journalists do not refer to any data collection process or analysis, nor are any of the contracts mentioned revealed. One claim is based on data unearthed in a “similar contract”, whereby the journalists claim that “we have been told”, demonstrating their reliance on their anonymous sources or Martin Wells.
Other quantitative claims seem to be based on hearsay: “Lord Bell told the Sunday Times”. Formulations such as “the bulk of the money was for costs such as production and distribution”, points towards missing data, or evidence for their investigation. Further sentences point towards numbers and use quantifiable words such as “a tide of violence”. Other facts mentioned still seem to be reliant on the numbers are given by Martin Wells: “five suicide bomb attacks”, which does not refer to any specific data or any evidence of such an event occurring.
Another point of criticism is that the numbers mentioned in the story are there for illustration or ‘evidence’, however the real story could be understood without them, for example saying an enormous amount of money instead of $500 million, would prove to have the same effect on the overall presentation and flow of the journalistic story. Furthermore, the reader is left to assume that the numbers mentioned are factual, and lack a thorough context. Such statements are easy to follow and if based on one real-life account, namely that of Martin Wells, lead us to believe that it is necessary to remain critical of such a journalistic ‘investigation’.
In terms of visual proof, the article does not provide much either. It features a 10-minute interview with Martin Wells, alongside some personal photos of his of his time in Iraq. However, the remaining visuals are just general photos, found through Getty Images.
Upon further investigation into the works of the journalist, it would seem that there is a strong preference for writing narrative-based journalistic works and focusing less on numbers. It is possible to speculate that the journalists, Crofton Black, and Abigail Fielding-Smith, have a preference for writing in such style, as the topics dealt with are incredibly sensitive, and involve several governmental institutions, the army, as well as other institutions who are not easily criticized and whose data is not available to the public. In terms of improving this specific article of investigative journalism, a strong preference should have been made on having multiple interviewees, in order to support Martin Wells argument. However, one of our personal limitations includes the fact that our criticism is based solely on this article, which is meant to be part of a larger investigative story on Privatised War by The Bureau of Investigative Journalism.
Essentially, it can be argued that this format of investigative citizen journalism cannot easily be compared to other works of ‘data journalism’, considering the lack of transparency in methodology, research, and analysis.
Link to the article: Fake news and false flags