 There are other questions which I can take offline maybe. Yes, thank you very much. Thank you very much. So the next seminar will be by Walter Quatoshocki. And he will again talk about social media, and in particular, eco chambers in social media. So Walter, can you share this? OK, very good. Can you see it? Yes. Yes, so it's all yours. Go ahead. Fantastic. So I will give a brief sketch, first of all, on about the approach we use to do in my lab. I mean, I'm at the University of Venice and leading the laboratory of data science and complexity. And basically, what we love to do is a data-driven modeling of complex system. With the complex system, you are just, I think, more in this kind of stuff, more than me for sure. However, it's a way to model processes in a way that everything is grounded upon interactions. And this can be used to afford transportation problem, biological models, economics, social science phenomena, and so on. Furthermore, in the last five, six years, we get this data delusion. We are overwhelmed by data. We have tons of data. And all this data, if implemented in the complexity paradigm, can be exploited to analyze and to understand social, economics, biological, epidemiological processes, as we saw even for the epidemics this year. The idea is not new. I mean, it's just, already in 2009, there was a science paper by David Lazar that said that requests proposed to join forces between complexity, social science, and computational computer science in order to bring up this new idea of the computational social science. So passed from a system in which we move mainly by speculation or a smaller experiment, passing on a more hard version of the social science investigation. Furthermore, this issue is particularly solid, even on the light of the problems in the reproducibility of experiments coming from psychology and sociology. I mean, there is this study in 2015 that was telling that almost half of the studies from sociology and psychology lack in experimental reproducibility. So it's really difficult to have a tough result and advance science with this kind of uncertainty in the kind of results we get. And so, after the proposal of David Lazar of the computational social science, several researches and investigators start to work on this idea of exploiting data from social media, one of the issue on social media, to understand something about the social dynamics. But as you can see on the top left corner, we have the distribution of languages by looking on Twitter on the state of New Jersey. So there is a new way of building a ethnographic map. You can see the location of people, you know where they are, and so you can recognize the languages so you can map the distribution of the place according to the language of the tweets. Or in the bottom right corner here, we have the distribution of the occupy movement according to Facebook. So according to the location for the people where they are meeting and so on, according to their posts, we built up the map of the distribution of the occupy movement on a global scale. However, my point here is that this is still not science. I mean, here we are in the realm of pure observational statistics. We are just describing a process. I mean, this is just a picture. We don't know anything about the process. We don't know anything about the dynamics. Today is like this, tomorrow it can be changed. So it is just a glimpse of something. It's a nice picture, but it is still not science. So in order to advance this issue, we framed this approach in which we started to grab information from various fields to set up a nice research question. And we started by the idea from the science communication of the agenda setting. The agenda setting is a theory that basically states that the more a news is covered by the media, the more it is perceived to be important by the users. So frequently and prominent discussion about the same topic get the effect to be perceived like more prominent and important to the audience. And basically this phenomenon is slightly changed in the last 20, 30 years with the rise of social media, where we passed from a system in which information were selected by experts selected by journalists. And so they select the information that have to be reported on the audience to a more hybrid system in which we have kind of this intermediation where the likes and comments actually drive the actual agenda setting people on social media decide what is important and what is not. And Mark Z just say that really innocently say that we don't want to build a new community. We want just to allow people to share and get information easier. And actually this utility component is actually the killer of the actual mainstream information system because when everyone select the information, everyone produce the information, the problem is on the quality of the information. And so we decided to advance this scientific research question by creating a kind of provusing, a kind of methodological approach that consists of finding research, framing in a good way the research question by studying the theories from other fields and then apply the classical, the class. We can basically show how we can answer to some of these questions by looking at the data we already have and the answer has to be sound and tight. So we pass from a part of data collection and transformation, we perform the quantitative analysis and then there is the modeling and validation part. So the classical scientific method. In the focus was on the spreading of information online, especially looking to social contagion, the collective framing of narratives, content consumption and especially on opinion dynamics, but on a data-driven perspective. So the question for us was is confirmation bias so the tendency to acquire information that are there to our system of belief dominant in the diffusion of information. And we set up an experiment by looking at Facebook pages in Italy and in the US. Basically, we pick two different sample of clusters of information providers. One is the conspiracy thinking, the one related to chemtrails, blah, blah, so all conspiracy stuff. And as a control, we use the scientific information sources. So pages that are devoted to disseminate scientific results. We divided these two datasets and we collected in Italy around two millions of users in the US around 55 millions of users. And the idea was to understand how exposure to a specific narrative, so conspiracy or scientific narrative affect the way we consume information and we use our confirmation bias. The first finding that was not surprising but really surprising, it was that people that follow scientific information source just follow only scientific information sources while people that follow conspiracy information just follow conspiracy information. So we get stuck in the kind of narrative we have, we love. Furthermore, we are surrounded by people doing the same. So if I'm a guy really active in my consuming scientific information, there is a high probability that in my social network, my virtual friends are guys and girls really active in consuming scientific information. So we create kind of homophilic clusters in the way we share and select information. And the experiment was really simple. How guys usually exposed to conspiracy information react to intentional joke made to mock their narrative. So if you believe in aliens, how do you answer to an information that says that aliens forbid to use soap or to its soap. So creating this kind of fictional mocking of the conspiracy narrative to see how those false information percolate within the conspiracy cluster. And actually the result is that the most of the interaction is come from people from the conspiracy side both in liking and commenting. The second result, the second experiment, the second part of the experiment was to measure how users generally exposed to conspiracy information respond to corrections. So if you believe in something that we know that is false, how do you react to a post telling that you are not right, that you are believing a fake story. So the banking, actually. And actually the result was quite depressing in a sense that they, the conspiracy guys generally ignore the sending information coherently with the hypothesis of confirmation by and dominating the process. And actually the small fraction of a conspiracy guy that interacted with this correction post after the exposure to the correction became more active in consuming conspiracy information. It is called the backfire effect. If you tell to someone that he is not, he is not right. Actually his reaction is gonna be to be more, to die in more deep in his kind of information sources, looking for his comfort zone. Now actually, let's summarize what this experiment tell us, the experiment says that we are surrounded by information so we can select actually the information we like the most because the offer is really high. We pick the information we like the most. We find people that share with us the same tendency, the same attitude, and we create this kind of homogeneous clusters. And the information circulate within this homogeneous cluster if the information adheres to the shared narrative of the group. So keys can be exploited to create models, exact models of information propagation online. In fact, indeed, the majority of cascades pass from nodes having the same polarity, so nodes having a similar attitude. Both for science and conspiracy despite the cascades, dynamics may result different. The process of capturing the information directly from a friend is a friend that is sharing the same attitude, this pretty similar. And there is the last final element. How, you know, there is this hypothesis that is called the group polarization hypothesis that states that when two users, extreme users, are in the same group and surrounded with users with a different attitude, these two friends sharing the same narrative will tend to be more extreme in professing their shared narrative. So like-mindedness create kind of a coalition in diffusing the information. And we measure how this phenomenon reflects on the sentiment expressed in the comment of the users. So the more you comment, how does it change the emotion that you are gonna express in your commenting behavior? And what we found is that the more you comment, the negative, the more negative are the comments bought in both the echo chambers. And it is even worse when discussion come between the users of two different coalitions. So if a conspiracy guys needs scientific guys and start to discuss the process became really, really harsh in a really small fraction of time. Now, of course, the objections we received to the study on the conspiracy science story was that polarization is a normal component imposed by the selection we performed. I mean, we select two different words, science and conspiracy. It is normal that people do not compel to this so distant narratives. So we keep the challenge going to see if this polarization dynamics nominate even the consumption of information. And we performed these analysis on 376 millions of Facebook users. And we analyzed how news get consumed on Facebook. And what we found is that actually are dynamics that is supportive for the confirmation bias I thought is. So the more you are active, the less. The number of news outlets are used as follow. So we concentrate our activity on specific page. Those is called selective exposure. And so we find a strong polarization, strong polarization that is consistent in several situations. We find polarization, looking at these, the same polarization that we find in science and conspiracy. We find it in the Brexit side. We find in the vaccination debate. And so we decided to explore the policies of the interplay between polarization and misinformation dynamics. So the more there is polarization, the more it's gonna be the probability that you are gonna meet some fake news. I don't like really the term of fake news because everything can be fake under a certain level of uncertainty. But let's use this because it's actually the mainstream definition. Basically, the idea is that fake news diffuses when there is a polarizing topic. So we set up a kind of framework. Framework, let's say a combination of algorithm for the analysis, which by accepting from the new sources and applying natural language processing, topic extraction and performing, combining sentiment analysis and some feature, accounting for some feature from the user's behavior, we can perform a classification in which we can filter potential misinformation targets according to misinformation dynamics. And as a final point, as a final remark, we think that this polarization, so this kind of clustering in eco chambers is a component effect of the social attitude of the users and the algorithm of the platform. So we decided to perform a comparative analysis on Twitter, Reddit, Facebook, and Gab to see if the tendency to be surrounded by users having an homophilic attitude is dominant and actually find that Gab and Facebook and Reddit provide this issue. However, if you look at the picture on the right, if we go to look on the polarization dynamics on Facebook and Reddit on the news consumption, we find that actually on Facebook, polarization is really dominating the process while on Reddit it is not. It is not. And the only difference between the two is that on Facebook, you have social feedback algorithm while on Reddit, we don't have it. And one second, this is the group that perform on this experimentation and they still performing this experimentation. It's really multidisciplinary. It's composed by computer scientists, people from economics, PCCist, mathematician, engineers. And this is my actual group working here about the impact of our research. And we get, for instance, the result on the banking that doesn't work and make and change the mind of conspiracy guys create and they will make it close the banking column of the Washington Post in 2015. So there was a column that was fake on the internet this week that ran the article why this is the final column and was siding the, just the plot I showed you before in the backfire of the backfire effect. We got a lot of media coverage. We get translated in all languages. We have one article on Scientific American. Actually, I wrote an update of this article that is gonna be out on next September. So how the information about the co-chambers changed over this issue. And our result, the result I presented here to you was used to inform the global risk report of 2016 and 2017. And actually we are working together with the minister of innovation, the Italian authority for communication and for Facebook even during the COVID time, what we did. What we did, we used the mobility data from Facebook. Mobility data, so on your mobile phone, users with their mobile phone running Facebook and with the GPS open are tracked by Facebook according to their position. What we found is that according to the mobility restriction implemented by several places from the various countries we focused especially on Italy and we found that actually the municipality that are more affected by the reduction of the mobility structure and according to the efficiency of the connectivity network are the most poor municipality in the countries. So we get that the impact of the lockdown and the mobility restriction is asymmetric according to the economic predisposition of the country. And this opens another question. How a nation is resilient to kind of mobility restriction? So we model the lockdown, the mobility restriction like a natural disaster. So there is a moment in which you are basically stopping your activity. And we measure according to this mobility data how France, Italy and UK responded in the implementation of the mobility restriction. And what we found actually is that basically the main effect, the really simple effect of the lockdown is a shortening of the small world behavior. So we tend to isolate clusters. So the reaction to the pandemic is to... However, this reaction is implemented differently in all the three countries depending on the previous infrastructure of mobility they have before. And actually it reflects the historical background of the nation. And we map and measure the resilience of this structure to a percolation approach. And you know, Italians are not so proud of their country, usually. I mean, we don't like, we generally feel really sad. We are Italians, we are not good at all. We are not good at all. However, we found that we have a really strong resilient network and it makes me happy because at least my confirmation bias sometimes is not confirmed at all in a positive way. And that's all my stuff. Thank you very much. Thank you very much for this very nice presentation. So I had a general question just to start. So do you have... Did you also see whether this phenomenon, this polarization phenomenon is getting better or worse across time? I mean, do you see trends in... It's increasing, it's increasing. Actually it's increasing. It's really difficult to monitor a trend because the number of people changed, the use changed, the topic changed. However, it seems that the trend is going to increase. This polarization trend is increasing. I see. Thank you. Yep. Chang-su, a very interesting talk. You just mentioned the evolution of the development of polarization. I want to have a question about this possible understanding of the dynamic of polarization. How... See when you have your experiment, how when the signal is propagating, it's just because you have underlying like a group of people, they already form like a dense community or during the process, you have some kind of reinforced like a dynamical process, for example, by recommendation, right? You have this kind of infection and spreading in a kind of explosive way due to recommendation. I don't think that social dynamics, information spreading can be more than like a virus, just for a simple thought. Yeah, yeah. When you select an information, you have the intention to share something. So you can select what you want and what you want. While with the virus, you cannot select. And so this process is dominated mainly by the homogeneity of the cluster. So you are being exposed to similar information. So you are not exposed to information that are different from your perspective, but because you don't want this to happen. And this dominates the propagation process, at least from what we saw from this data. I mean, I'm not saying that this is universal. From this data on Facebook and with Facebook, Twitter and YouTube and Gabbo, we can see this actually. Yeah, so the people try to spread, they already know the targets where to spread. Yeah, yeah, thank you. So there is a question from Leihang. Maybe Leihang, you can ask this question yourself. Okay, yeah, just from this slide, there seems to be significant difference between the Northern and Southern Italy. Yeah. What are the ramifications in terms of the things you discussed in this talk? I mean, because the economic, as I told you in the previous paper, the last Pina yes paper we get is on the Italy. Basically, we found that the main mobility restriction affect the poorest municipalities. And actually, the Southern part of Italy is the one with the finding difficulties in the development and economical development. And this is given even from the infrastructural connection. So railway station, trains and so on are really not well connected. So we leave since a lot of time, there's really differentiation in areas that reflects actually really our historical background. Okay, so both economic and cultural. I think more economical than cultural. I think more economical than cultural. Thank you. Okay, thank you very much. So I think we extended a little bit into the coffee break or break and so if there are no other questions, then I think we can stop here and talk all the speakers for a very nice, oh wait, I think there is a, looks like there is a question. Do the echo chambers network shows small world characteristics? Yes and no. Those are really dense and connected community with a short diameter. But it was more like a globally coupled. Sorry? Sorry, okay. Why with the short diameter, but the overall Facebook is with a short diameter. So all the social media software is a small world effect. So I don't think that the, I mean, actually it already happens but it's by definition is the shrinking diameter of the interaction. Okay, thank you. So, okay, so we reconvene at half past 10 Central European time and what time will it be? Time is 4.30 PM. 4.30 PM, yeah, okay. Okay, thank you. Thank you very much again and see you in a while.