 So, let me say a few words about this project within the MariaMIF tool. So this is about social interaction in online participation platforms and let me put things in context. So these are generally online spaces where many to many social interaction occurs about common topic of interest. This is a rather general definition. It can include news aggregators, websites where we talk about Brexit news or a Wikipedia website where authors jointly develop an article or it can also be, I don't know, these massive online courses where the students discuss about certain topic of the course. So a prominent example of this type of participation platform is open democracy. Open democracy is receiving increasing relevance and attention. So for example, there is this section in the Horizon 2020 that supports for funding and there is this section, this report of collective awareness platforms for sustainability and social innovation. So there are also some successful initiatives in Finland and other countries and this is attracting a lot of attention. In particular in our country in Spain, there has been two recent initiatives on that side. One in the City Council of Madrid, which is the field in madrid.es. You can go to the webpage. I will mention a bit briefly the structure of these websites. And in the City Council of Barcelona, more recently the City in Barcelona. So for us as scientists and researchers, I like to see this from a point of view to nice research opportunities. So the first one is that these platforms gather continuous data from complex social interaction and they are challenging in the sense that they require us to develop novel computational tools and novel statistical methods in order to better understand what's going on in these web spaces. But not only that, it's because since we have some type of collaboration with the people behind these platforms, we can suggest changes in the platforms and then evaluate how these changes affect in the social community. So there is an option there to intervene, to make social experimentation and this is not just a data set that is offline that you can download or do some offline analysis, it's something that is evolving in time and it's changing and allows us to do different type of analysis. So just some few slides that explain a bit these spaces. So the field in Madrid was, as I said, created in September of 2015. There are basically three sections, participation, transparency and open data. We focus more on the participation side and here we can, these are some numbers, so far there are 150,000 users, so you need to be registered in the community of Madrid in order to participate in these debates and proposals. There are 4,500 debates, 2,000 proposals and this is growing in time. So when you look at the participation, you can see these two types of things. So on one hand you have the proposals where registered users can propose something. So here it's hard to see that these are the most three successful proposals that have received more supports until now. Okay, it's difficult to read but one talks about removing bullfighting from Madrid. The other one talks about some tax to the church goods and so on. So these proposals are generated and then they are discussed by the community and if they are approved because they reach a threshold of support, then they go to the parliament and they are discussed there and they are forced to be approved in the case that they are being approved here. Of course, so far this is very new, no proposal has been accepted yet, so it is far from reaching that level but this is the way. So apart from the proposals, we have the debates. So in the debates, for example, this is a debate where you see here is a conversation between people around a certain topic, this is the example of the debate of the bullfighting case and you can see that people reply to comments and reply to replies of comments. There are some comments which are very short, there are others that are very elaborated so some people really take time on writing on these type of platforms. And on the other hand, the city in Barcelona which is more recent was created in February 2016. It is actually based on the same type of source code so it is a fork from the same repository but the group is different, it is here in Barcelona and they have different types of proposals. There is also not only the citizen proposals but there are also proposals from the city council, from civil organisations, there are some periodic of regular meetings and they add proposals to that. So this is an example of the platform with some proposals and other supports. The mechanism is very similar in both. So what do we do with that? So this is a collaboration between the UPF, the AI group and mainly the Eurekat with a group of digital humanities, led by Andreas Kaltenbruner and there is Pablo Aragon who is a PhD student of ours. And he is basically the person in contact with the organisation of these two platforms in the two different city councils. So he can go there, they have regular meetings, he suggests changes to do in order to improve the platform and then this type of dynamics. So far what we have done in the context mainly of this European project that we are now extending or adapting for the Maria Máez II is to develop open tools to facilitate some activity monitoring or some that can help us to do some global statistical analysis of the activity in these spaces and how can we assess better citizen participation. So these tools have already had a big impact in terms of dissemination activities. For example, in the platform society conference or democratic city events, TICTEC conference in Barcelona, the big language exhibition. So we are quite surprised about the visibility and the impact this is having in society already. So this is an example of the tool that has been developed by Pablo. So here you can get a global picture of the activity that is happening in this website. So for example, this is for the city in Barcelona. You can see the number of comments and the number of contributions. In times you can see peaks when this thing got to the public media and you can see here distribution of the different people that contribute to the proposals. Some charts and some tech clouds. So the nice thing of this visualization tool is that it can be adapted easily to different sources of data. So for the City Council of Madrid, you can also use this tool. And for other news aggregators, as we have done, you can also use it. And this is very intuitive to get a first understanding of the activity behind the platform. So another tool that has been very successful so far is this idea of representing a discussion thread not only as we saw before as a plain interface or a threaded conversation, but to have really this type of rooted radial tree that grows in time. So this is very useful in order to get a first impression of how the discussion is being developed. So by browsing through the different notes, it's not very clear here, but you can see the different comments of the people. You can populate this growing graph with some attributes like the degree of controversy of a given comment or the number of the size of the notes and things like that. So this has been very successful and I guess it's going to be adapted in the City of Madrid. It's going to be implemented in this platform soon. So this is about visualization, understanding more or less the social interaction that is going on in this website. What we are also another line of research that we are pursuing is how to measure changes in the interface, how significant a change that we want to do, how affects the social community behind it. It creates some significant changes or not. So this is this type of line of research. So here we focus in another type of platform. This is not the City Council, but this is Meneme, which is the most popular Spanish social news networking. So there's a lot of activity in this website. I don't know if you have participated in these forums, but they are really huge. They have millions of posts and comments and every day there are more news that are being discussed. And this website underwent a change in the interface. So it went from showing the discussions on a plain interface where you could link to previous comments by just tagging them to a type of interface where you can show the discussions in a threaded view. So this was an opportunity to test this type of controlled changes. How can we measure how they influence in the productivity of the website or in the performance? So we are doing some type of statistical analysis on this website. And these are the examples of results that we have. So you can see that the discussions, this would be a discussion before the change in the interface where you can see that most of the comments go to the root of the news post and then there are these big chains of two users that enter in a fight or a discussion where they basically insult themselves or they basically create a very constructive dialogue. On the contrary, after the change interface we can see that the complexity of the dialogue and the debate is higher and we can see that there are structures that are more complex in the sense that they have different values for some of the measures that we try to quantify. So this is an example of such a measure which is an H-index. The H-index you can use to characterize the complexity of a discussion. And we developed methods based on regression discontinuity analysis or the standard statistical tools to detect these type of changes. So here we can see that after the change in the interface we can see a very significant change in the complexity of these debates and discussions. So each point here is a conversation or a group of them. So far I just talked about visualizing how can we get an overview of the activity, how can we identify changes or how can we measure changes in the interface. Another type of research that we are pursuing is how can we define changes in the interface in order to optimize some performance. So an example of that, again with these type of discussions, would be for example if we suggest a comment, can we change the shape of the discussion in a certain way that we think it's more interesting. So for example think about this measure that I was showing here, can we optimize some type of structural measure of this conversation and influence the social dynamics in that way. So this is some work that we are doing with the people in Naimechen with the machine learning group. And here the idea is to use some type of optimal control in order to optimize this type of structures. So by suggesting a comment we can influence the, so this is simulated that of course it's not real data because we don't have the possibility to make such an intervention but this to show theoretically this type of approaches can be interesting in order to optimize this type of performance. So here you see conversations that are this one is a real conversation, this one would be a conversation for the model estimated from the data and this one would be a conversation generated by the controlled model. Here the h-index is bigger to show that this can be used to optimize some type of structure. And this something general it doesn't need to be applied only in discussion threads but in general growing networks. Okay so I talk about conversations but I didn't say anything about the content of conversations and of course if we are talking about online democracy and things like that the content, the text there is very important. So that's why we are collaborating with the Natural Language Group, Natural Language Processing Group here at the UPS with Horacio and the two Francescos and they are looking at opinion mining and sentiment analysis at the level of these comments. So this can help us to better understand things that are occurring in the website. They also have this study done on these emojis on Twitter, so we are now adapting it to apply to many of them as well. And so I mentioned this roughly these lines of research but there are many more so for example if when one enters into a discussion a good tool that it would be useful to have is a summary of what has been the discussion so far. So summarizing debates can be another interesting line of research or how can we detect argumentation between different people. And for example in the city council and platforms there are these proposals generated all the time. Some of them are very similar and it makes sense to group them in one single proposal. So tools for automatizing this procedure can be very useful. Now more from a social complex networks analysis one can be interested in understanding how polarization changes in time as the discussion evolves or some proposal evolves and how can we detect communities within this type of debate. So for example in the case of the bullfighting where we detected some group that was performing some lobby against bullfighting, some other group that was in favor of bullfighting, some other people that was not belonging to any lobby. So there is a complex hierarchical community structure in these social networks that we only get to see some biased samples from that and we can work on methods to try to identify these communities from the debates. And of course one can think about recommendation systems so suggesting comments, suggesting proposals, all these things are very necessary and we are working on that. And many more. And this is just to summarize what we have done so far. So these are the main collaborations, the artificial intelligence group with Eurekat, with the natural language processing group, with the people at Naimehen. So of course this has a lot of relation with what Victoria was selling avant. So we cannot just publish the data on the web with all the information of all the people of the city council. So we didn't release this data yet. We are in close contact with the city councils and see what are the best ways for doing that. But we really believe that publishing at least the tools can have a big impact. So for example we just released this data set that contains some discussions from Slashdot and other websites that we actually did some research years ago and just putting it there I started to receive emails and there were a lot of visits that people that were interested in in doing some type of social analysis with this type of data. This is also the visualization tools I was mentioning before. And well to mention that Pablo won this award for the epic PhD workshop. And maybe this is a bit off topic. So maybe we should have said this in the talk of yesterday, but this is some workshop that we are organizing collocated before NIPS where these topics are also marginally related. And I think I'm done. Thanks. It's not in such a good shape, okay? So not only in Spain. And of course the question is whether these type of tools can help or they can provide sort of a cover up for the current deficits. That's one issue, one question maybe, I mean that I know it's not technical, but it's a very interesting and relevant issue. So what is the impact and the quality of the democratic debate look at this type of debate. So go, this is the land of the trolls. So we are slowly becoming the troll society, troll candidates running for president and things like that. Okay, so that's one thing. The other thing is maybe natural language, okay? It's not the right language for constructive discussions. Maybe with this approach one can come up with some other tools. Let's say graphs, okay? So you're talking about something. You build graphs with support and re-battles, okay? You express sort of graphically so that people can in some sense grasp the complexity of the issues and people can collaborate to elaborate these graphs of supports and so on. So whether you're looking into something like that as well. Yeah, well, so regarding the first question, I don't think I have much to say. I can say that I believe that these type of tools are good and they make a good service if they are well done. So if there are a lot of spammers and a lot of trolls, you might think of incorporating some utility that makes a spun filtering or some mechanism in order to avoid that. So I guess that the answer is yes. It can definitely help if you do it properly. And for the second, yeah, that's definitely very, very what I was thinking when we were talking about creating these types of summaries. It doesn't need to be natural language. We actually have this radial tree representation to represent the discussion. So instead of using the comments as the nodes, you can use some arguments or some more compact representation of the different argumentations of the debate. But this is definitely to look for a different alternative representation that can be more suitable to understand the discussion. This is a very interesting topic that we are working on that. My question is... The first one. I have two questions. The first one is actually related to what Hector asked and actually your answer, okay? And is how do we know that you are not manipulating democracy by the tools that you are using? For example, the answer that you gave about trolls, who decide who is a troll? I know this is one that has to be somehow you have to demonstrate somehow in your results, I guess, opening your core, explaining how you do it, and see if that... And the other question is really one of the problems with these kinds of situations is this is discussions that people do in a bar. And that's in reality, life doesn't work like that. It doesn't work like the discussions that you can have about a football game. There are other practical issues that people ignore, and this tool could bring into the discussion automatically, like costs, like population numbers, something that usually supports or helps somehow people to visualize what is the reality of the discussion that they are having. Okay, so is this a possibility of just somehow using these to bring the actual data that exists about the topic that is being discussed? So actually, the second question is very related with what Hector was saying, so these are alternative ways of representing the underlying interaction in a way that helps us to understand better what's going on and in order to facilitate better performance overall. So I think that this is possible. We need to work on the particular methodologies that we want to do. And regarding the first one, well, I am not a politician, I can only answer the same as I did before. If I want to prevent trolls and you want to see what is the method we use to prevent trolls, we can make explicit what is the strategy that we do for troll detection. There are several, for example, in Slashdot there is this score-based mechanism where people evaluate the comments that other people write and there is a filtering where the score is too low and this turns to work well in the sense that the spine is avoided to a certain extent and high-quality comments are rewarded. So we can think about this type of mechanism that if they are centralized, we can make them public so that people understand what they are really doing and if they are distributed and self-organized, the people can also understand what's going on. I don't know at this.