 Hi, welcome at the stream of Chaos Zone, which is basically a group of CCC groups in the east of Germany. It's a talk by Korrektiv and Clemens Kommerel. No filter for the right wing. This talk will be translated live. In your streams, just choose the translation section and then also that is coming right now. Have fun, enjoy the talk. No filter for the right wing, for the right wing people by Korrektiv, have fun. The picture is like a common Insta post. A white baby, a blonde white baby, a wooden sword, a little bit of decoration. The warm light looks good. This 669 people like it. Who looks at it whenever you're looking at it more closely, it means something different for them. This is a little kid child that will be responsible for, that will be ready for spring attacks. The wooden circle is supposed to be a black sun, so this is a small child next to a neo-Nazi motive. And welcome to the bad side of the platform. This is what we wanted to look at. So even after short research, we knew that this post is only one of many. When you're white, there's no upgrade. Don't mix. This is a world that really doesn't really attract us and we really have discussed when we're looking at this. We, that's a five people team from five reporters, Alice Echtermann, Till, Arne, Celia and me, Clemens, who is responsible for the data analysis. And then there's all of the Korrektiv team. Me, myself, this is my first data journalism project. And I'm really happy to be here representing the team to show you all our research, no filter for the right. This next 40 minutes is about Instagram and the right wing data. This is about how they are using Instagram. And it's also a report of how we took this data set of four and a half million Instagram users and how we researched it. Why Instagram? It's one of the most common apps currently on a first look. It looks really, really plain and not really dangerous. It looks like nature photo, photos, travel blocks and stuff. There's more than 20 million people on there, German people. And half of them are young people. So there's a lot of young people when they're, when they're looking for their identity and they really are touching with the right scene. And the number and the size of right and right extreme contents is larger than I expected. At least I personally thought so. I personally had never been in touch with right wing data, right wing content on Instagram instead of Twitter, where it is more across bubbles. If in Instagram there's only, you only stay within your own bubble. So I've showed you some, we have some examples here, like Twitter, like users, like politicians and even some posts that really look not very dangerous. There's some subtleties. Picture, for example, is from Lisa Lehmann, who looks at, who's in the Young Alternative Party, which is a right wing party in Germany. And this looks kind of not dangerous. A young person drinking her coffee and the, but the script says taking a break with coffee that is by someone. So it basically says, well, Germany is still occupied by alleged powers at the hashtags like photo, picture, winter, show that even when you're looking for not for harmless hashtags, you will find this picture. This is only one example of many, this, these examples are enough to show, to be able to map the network of the new right wing party. We knew that we needed many, much data for this. And this is what we want to start with. That's why we're first looking at what data you can actually get from your Instagram profile. In my opinion, that's four parts. First of all is like meta data and common profile information. So that's basically names. And for all of the posts that this person has made, there's data on like how many likes, how many comments, when was it published? And then there's hashtags. And some of the posts also have subtitles basically. The second part is connection data. So who follows this account and who does this account follow? The third thing is text data. So like the subtitles, the description, what's on the picture stuff like this. And the fourth is like subjective data, so qualitative data. What does this picture look like to me? What emotion does it convey? This helps to understand all of the other data. What is missing in this are the stories. These are only available for 24 hours. So we weren't able to look at them. But there's still lots and lots of data that we do have. So we've set up a database to basically save all of the data. This allowed us to also save changes in the data. But we also needed a sampling method because we wanted to trust our sampling. We couldn't just take some accounts randomly. Instead, what we did is exponential discriminative snowboard sampling. It sounds very fancy, but it is actually quite easy. We assume that each Instagram account follows other accounts. And then we follow these accounts. And then we follow these accounts and so on and so forth. In theory, we would get to a huge sample very quickly because we know since this year how exponential growth works exactly like the coronavirus. So we needed to basically stop at specific points. We basically just filtered some of these branches and we'll look at how exactly we did that. But first, we need origin accounts from which to start. So for that, we created a fake account. We basically made it look like a right-wing account, like it was part of the scene. So it would allow us to be in this scene. Within this, to select our origin accounts, we basically started following parties. So because all of these party people are paid by public money, so we thought it was very good to look at them. And then we looked at additional sum accounts within the identity movement in Germany that the Constitution has basically considered we need to watch them. And then we made sure that at least a second person within our team also looked at this account. Overall, we had 281 origin accounts in there. Most of these accounts came directly from AFD, which is the German right-wing party in parliament. So obviously it's not exactly a perfect sample. There's a good chance that we missed some other groups. In parallel to these origin accounts, we also included a control group that basically includes VIPs and influencers and some accounts from our environment. We wanted to also find some content that isn't only found within these right-wing bubbles. So let's look at what we did. First of all, we looked at all the connections from 281 accounts. We looked at those following those that these origin accounts were following. Overall, it was 85,000 profiles and many of them were only followed by one person. So we chose to use a criteria. So in order to stay in the sample, an account needed to be followed by at least three of our origin accounts. And that was 4,532 accounts. And then we followed additional accounts and then we had 800,000 accounts. So then we had a second criterion. We wanted to keep accounts that were important for our origin accounts. So every account became points for different connections. We basically gave them different points. And to stay in the sample, you needed to have a specific number of points. This number seems random and basically chosen by us. And it is. But we needed to basically balance having a sample size that we can work with and having enough coverage. And this seemed to be the best compromise. So we had 10,805 accounts overall. More than a third was private and we couldn't use it. And in the third step, we basically categorized these accounts into important or not important. Not only, we didn't only categorize into important or not important, but we also categorized into categories. We basically had a training using a sub-sample. Shows choosing these categories and then we categorized all of them. In order to easily make categories accessible, we basically made an online interface and every member of the team could basically look at the profile and then choose the category that it should go into. And in addition to that, there was the possibility of adding some comments. This is very interesting or part of the identity movement. You could also skip accounts if you couldn't find anything or give it to some other team member. Any account that we thought would be a control group or not very interesting. So we actually changed the weighting in our database and then they basically fell out of our sample, even though we didn't look at them specifically. Overall, we had more than 100 coding sessions with a lot of interactions. Overall, we spent 120 hours on Instagram profiles of the right scene. Because we got a feeling for the activities and their preferences of this scene, which we didn't know before. Six and a half thousand accounts were categorized. There were over eight and a half thousand relevant accounts and we have found four and a half thousand to be relevant. Some of them became irrelevant because the network changed and the dynamism removed some of their criteria. There were a bunch of accounts that we didn't categorize, three and a half thousand, because they were private. And then the dynamism caused us to miss some of them. All of these were close to the two and a half points that we were assigning. And this final set sample of four and a half thousand, we then went to fetch all relevant data to do our assessment. We have 4,501 Instagram accounts, including all information, profile, followers, posts and so on. Between those accounts, on the first of September 20, we finished, there were 330,000 connections. We have over 800,000 posts of those accounts. And finally, we have a category and a subcategory for each of the accounts. That's the second part of the presentation. Now we'll take statistics on how those accounts were categorized and then we look at one piece of data out of the metadata. Here's the split of those categories. The largest ones are AFD and Patriots and Conspiracy Theorists. When we look back to the origin where we came from, 80% were AFD, so this is very convincing. The subcultures is the next big one and that's right musicians and lifestyle merchandise with subcategories. Medium sized categories are student houses and so on and alternatives and fraternities. Very small amounts are what we call real hard right wing political parties and so on. We categorized that as a separate group. Now a large part of the AFD is context. They were somehow part of AFD but they didn't have a political mandate. And there's party organizations. Maybe Europe and some of them are fan accounts. And they would be asking for somebody to become chancellor or something. Another large part is the Patriots and Conspiracy Theorists. We were attributing them to the scene because they had emojis and flags. And topics like migration and refugees, corona deniers and so on. Young means up to 20 years, 18 years. We have seen many profiles that talk about clawing back the country. And I wasn't aware of this and I didn't expect it before I started this. So these young people, new German standard, NDS, they identify themselves with this and they follow people who explain this. There's also people who idolize history, the Kaiser and the German Reich. Now we want to look at metadata and what can be done with it. The hour of the day and the day of the week when those posts were appearing. There's the party organizations on the left-hand side, on the Y-axis. There's the day of the week and across is the hour of the day. The lighter it is, the more has been posted. So Monday through Friday there's lots of posts and they start early in the morning and sometimes in the afternoon. But after 1800 it gets less. So with the politicians we find less at the beginning of the week but we find it throughout the day and we find it also at night. So this is widely spread. We expect because politicians have different preferences and these accounts are covered by multiple participants. Now here's the media scene. That's the media scene. So they emphasize the daytime. They start at 10 and 11 in the morning. So in the media area they sleep in in the morning and they work later at night and they work predominantly during the week. Now we're looking at the network connections on how the accounts interact in a network fashion. How does network data work? So it has nodes. That's the four and a half thousand accounts which we have taken. So we have Beatrix von Storch, we have Jörg Meuthen and Alice Weidel. So the connections show who's following who and only Alice is not following Jörg. So this is a directed network. And a connection can be unilateral or bilateral. You can see that the arrows have a different width. They have a different strength. The stronger the edge is the more important it is. So Mr. Meuthen has a very strong connection to the network. So we think about how it is weighted. We've tried multiple things because there's no real rule. And we've thought about what can we actually use a criteria. We have normalized the data. So here's the things that we use to weight the edges following each other. That's an expression of interest. So the amount of commonly followed accounts. So if you follow the same people then you have similar interests and you belong to the same group. So if you use similar hashtags you again share a topic. How often do they comment and how much do they follow each other's pictures. These are factors for interaction. And the more frequently it happens the more probable it's that they know each other. So for each pair of nodes we have calculated one of those weights. So it counts with more commonality. More commonalities are closer to each other than other ones. So with these weights we have everything that we need. So here's the graphic that is generated from that data. The closer they are, the stronger the links the closer they are to each other and the further they are apart. We are using the commonality class community detection algorithm. This algorithm tries to maximize the number of internal connections and minimize the number of external connections. So these are the clusters found by the algorithm and we have found a good overlap. We could give them names easily. So being close to the network doesn't imply a hard closeness but it suggests it. So accounts from AFD covers the whole left side of it. On the right hand side you see identity politics and it's close to subcultures. So if you want to look who is interested in what then a different representation is more useful. So here's each node one category. The size represents the amount of the accounts in there. The strength of the connection is the number of existing connections divided by the number of possible connections. And then you weight it by the weight of the edges. So it shows the interest of one group in the other one. The color of these edges is who is being interested in order to keep it intelligible. So AFD has a strong connection to the young alternative. But it is well explained in real world. So the young alternative is quite interested in AFD and they are strongly interested in identity movement. That one is interested in young alternative but strongly in subcultures. We don't only see who is interested in who but we also see who is among the top four. All of them are connected to the media scene. It's surprising. It's a very interesting fact because we see that everybody is consuming content of the media scene. The media scene and the topics and redistributes it. And that's what you see here. We can't always only look at the whole thing but we can also look at individual accounts. Two values are important. Eigenvectors and between the centrality. So the first one is somebody is as important as the accounts that they know. The second one between the centrality is a measure of connectivity. What are the steps on the path between two accounts that aren't connected directly? So they are how are they in between other ones? So on the left hand side you find mostly AFD and Junger Freiheit. It has a relationship to reality. All of this is important in the Instagram network because they are being followed by many accounts. And by important accounts. More interesting is the other one because there's lesser known ones. So Dubrov Kramanlisch is the city council Freiburg and we haven't found that much. And there's the rapper prototype and he has inspired the NDS community. So now we're going to put this into context and that makes it interesting even more. So now we are finding the bridges between those sets. All of these blue ones are AFD and we see in gray we see the ones that follow at least one of the two and our connection. AFD is blue, red are the ones with a very high connectivity. So they have a very high between the centrality. And we've already seen those names in the list. Here we see the peri-petit shop. That's where the research and data go together. Peri-petit is a fashion shop, patriotic fashion. And the known celebrities are advertisers for that label. And he's as a politician he's a good fashion model. So this doesn't only show the So this doesn't only show so this brand is really within both politicians young AFD and young identity people. So it can serve as a sign that we can't actually see from the outside but they will recognize each other. And then I'll look at which AFD politicians are interested in identity in the identity movement. And these are those that are red here. All politicians in red here at least three accounts of these identity movement. Of course, following doesn't mean that they agree but it means having an interest. Especially these connections are especially interesting because the AFD said the AFD basically has an agreement that members of party cannot become part of the identity movement or of real right extremism. And we've asked Alice Weidel how they can follow these followers and she said well we can't say whether following these identity movement people against the agreement that we've made. So most of them are local politicians but some of them are actually within German states. Both of them are, for example, Daniel Freyja von Lützow and Deniz Hochloch. Especially Kaiser AFD is the most important account within the SAP network. And it's very interesting because we haven't really considered it before but it has been an important connection node within this. Kaiser was a model for the NRW which is a state in Germany campaign model for the AFD and she actually wants to apply for a list mandate for the federal parliament in Germany. She follows many accounts in Germany especially identity movement accounts also from the initiative 1%. She also doesn't really agree to the agreement that they wouldn't follow these right-wing extremist parties and she also modeled a clothing from the shop that we talked about before and we see that these connections that we see on the platform are also reflected in the reality but what's interesting is that she deleted all of the posts, also those that we can see here. By now, basically there's a couple new posts and she also started following some new accounts so being able to say that she keeps her distance to right extremist positions is not possible. Lastly, we want to look at hashtags. Hashtags are important to look at themes on Instagram because they are really interesting for the content. Here we see all of the hashtags that are really common within our sample but not our control groups and we have the number of accounts that use these hashtags. Fraternity, Merkel has to go. Let's elect AfD Fatherland and these are often used from the IFD but we also have these hashtags from fraternities that are very common here. It's quite interesting to us but we can also look at it because many people from this identity movement are actually part of these fraternities and many of these politicians were part of these fraternities and there's a couple of hashtags against left extremism or antifa and if you look at it you can have an idea of the themes of a specific part of our network. On the left side we have the most common hashtags of the AfD and you can see that they're really quite egoistical. There's lots of AfD, elect the AfD, choose AfD, AfD is important stuff like this. There's lots of narratives that we often see with AfD. We are against the establishment the only way for you to reclaim your country is with us. If it wouldn't be that bad it would be quite boring actually because it's what we always hear. On the right hand side we have the patriots and conspiracy theories we have more of a focus on Germany like Vaterland, Poison, which is a part of Germany and is often considered an old part and what's important is that there's actually Defend Europe is quite present here which is part of the identity movement and it's basically against the EU and against the EU taking refugees as well and we can also see that it's often about on history and on the German Empire and then there's the hashtag NDS and there's these right wrap scene but what you can also do with these hashtags is look at how often are they present together? Obviously we could only look at some so how many posts were these hashtags together? In order to be shown here it needed to be at least it needed to be relevant for the end of the accounts and then we were able to filter it a little bit and then we can see sort of like a thematic map of the themes of the right wing in Germany on Instagram so we have in blue all of these different AFD type things then on the left side of that in the darker blue we have stuff that is which is the Austrian equivalent to the AFD and this is more about feeling home, keeping home but the AFD, the German community usually has the word fatherland so they use different words apparently and then there's a couple of English speaking hashtags and some within the identity movement that are pink or yellow and there's a couple of classical Instagram hashtags like casual style and follow that are also here but they're not as important and the last big cluster is hashtag is related to the hashtag so fatherland and it's mostly related to fraternities so we've seen a lot of it looking at data but there's a lot of stuff that wasn't measurable that you can't count where you can't have numbers, we ask people we interview people and these are also at least as important as the numbers so something that we thought was quite interesting because around the Black Tuesday almost all of the left-wing parties basically used this black square but then there was at the same time this white square and there were a couple of conspiracy theories that there are only 6% white people on the world anymore and these were parallel to the hashtag Black Lives Matter, they used the hashtag White Lives Matter we have this classical idea that right-wing people have we are being replaced and that's only one example of many and we have five stories that we've made that you can read on they show how important young women are for this movement how important some of these meme accounts are now let's get back to the beginning this picture with the black sun was unlined for more than 11 months only when we told them only when we told Instagram about it they deleted it according to Instagram, 350 people are responsible for deleting these extremist ideas and we want to ask them to basically follow these hashtags and to investigate these hashtags now apparently one of the hashtags that has been deleted for a while is still quite common in our data Instagram also has told us to please tell them more accounts, please send more accounts to them now working filters against right-wing really look different it's important to actually report these accounts and these pictures and we need a lot better media literacy we need to go into the schools what are codes of this scene what do these apparently harmless hashtags actually look like now I can only say thank you for listening thank you for being here I would recommend that you read these stories on Keinfelter für Rechts I'm happy to ask them to reply to questions you can also tweet at me at klickom thank you for listening thank you for this nice talk and we have this possibility of asking some questions because Clemens is here live on the stage and there's a couple of questions how do you define the right connotation of a symbol is there an overview, how do you define that okay, nice good evening, the symbol is we have a lot of different emojis we have collected them like the old German flag that is like an emoji for the Reichskrieg flag so different emoji combinations and some combinations of animals like boars and with a little bit of research you can actually find them and understand them even though you don't actually have them at the beginning a next question that might be a follow up actually do you have did you find a new aesthetic so to speak so is there a right wing aesthetic of Instagram that is different from how Instagram usually looks I wouldn't say so, no I would say the dangerous thing is more that it's really similar to really classical Instagram accounts it could be on every mainstream Instagram timeline there are normal hashtags are used and then they add some additional scene specific hashtags of course there's some meme sites they're very very close to these AFD look, there's some of them they're really common in the community another question also from the internet from the software that you are using you've mentioned Gaffi in your presentation, what software have you used to get to these slides to make this possible I'm not allowed to say anything about how we got the data we saved the data in a Postgres instance on a Ubuntu instance and the front end is a little bit of JavaScript and Node.js API that I built that in the database we used conjobs to update this and for the analysis we used Python, everything that's not Gaffi and we used Gaffi for this and we used Gaffi for the network data and then the slides are keynotes did you also look at likes and comments for the interactions comments yes, likes did not follow in there because we were unable to collect them there were way too many of them to collect them so even with software and maybe at the last question also up to date what about the relationship to the Querdenker which are the corona deniers in Germany currently, it would be really interesting if we had current data we have found we have found different Querdenken accounts especially like patriots and conspiracy theorists we were considering the Querdenken theme but it didn't really spring to mind ok then thank you so much for your presentation and all of the correctives theme we would be excited if you would stay with us at Kajasund Design TV the next talk is DeafOps and we from the translation teams are saying thank you for listening to the English translation of