 Welcome back from the break. We talked already about the role of algorithms in social media today in the panel on media and education. And yesterday we already had a discussion in the context of the roots for a democratic and peaceful society. And then we also had some site discussions about narratives around protests and around social media discussions and how they play a role today. And now we have a talk about peace building in the digital era with Andrew Suthayu and Dava Abdullah. And I give you the floor for the talk now. It will be about 35 minutes and also that we can have 10, 15 minutes Q&A afterwards. Yeah, give a warm applause to the reference. Thanks. Thank you very much for the introduction. Hi everybody. My name is Dava Muhammad. I'm a journalist working for a local media organization called Curson24 based in Iraqi Kurdistan region. Today we will be presenting to me it's a really interesting topic and also a very new one as I have jumped into this world almost very recently. And the reason that I really wanted to take part in this was to really encourage people and also deliver and convey the significance of peace building efforts digitally. It has been a topic for a very long time. But as with the rapid growth of technology and digital media, we are more in need of this critical tool in our lives. So today I'll be talking about peace and conflict and digital era with my colleague Andrew who will be presenting the technical part of a very important tool that will be helpful in terms of achieving the end goal which is achieving peace and building peace digital world. I'll go throughout my presentation. I'll talk about these two concepts as well as my very recent participation in an interesting research project. It was called D-Maps organized and presented by BuiltUp which is a non-governmental organization and funded by British Council. Mr. Andrew was one of the amazing member of the technical team who developed let's say a tool system that really helped us as digital peace builders in terms of identifying all the issues and all as well as designing interventions to resolve them. So I will be presenting my part and then Mr. Andrew will walk you through all the technicalities of that tool that was really helpful. So firstly peace and conflicts have been the almost the two most long-standing themes and marks in our lives in everyone in everyone's societies around the world. It depends where you leave I get that for example I am from a region the Middle East which is known and which is known for its long-standing conflicts whether it's religious ethnic sectarian they have been there because it's a really what we call it a hordes area that in terms of climate as well as in terms of you know ongoing conflicts sometimes they get really violent and this is what really concerns people both inside the region and outside the region as they depend on the region for a lot of its resources and and the connections and the geographic similarity. So it's not new that's my message peace and conflict are not nothing new that we are discussing today what what is new to us or what has become a new trend is talking and discussing these concepts in the digital world we are very invested in technology and its advancement but unfortunately little attention has been given to all these consequences that are the results from this technological advancement peace and conflict are you know two parts of this of this issue that have been given less attention so if we define peace I think anyone can have a definition of its own but the definition that I picked up and I believe this that really fits the concept of peace is it's not only about the absence of violence it's not the it's it's not enough for you the the weapons to seize where it's more than that it's a climate it's an atmosphere in societies that people feel they they can thrive they can develop themselves they can grow that's peace it's it's it's it's a kind of situation where you feel secure where you can predict a foreseeable future and you feel you have the sense that this is gonna be the case for a long time on the other hand we have conflict which is basically it's a clash of opposing feelings or needs that's very simple we see that conflict in every aspect of our lives as we see later so what is the difference that that that's that's gonna be the next question and I know that many of you might have it in your minds what's the difference between this digital this peace building effort in real life and digitally and that's an important question and I think that's the main theme of this talk that I'm giving right now the there are similarities for sure but the the differences are the differences are the things that make this effort unique and requires our special attention in real life we see that diplomats politicians researchers university lecturers academy from the academic community generally come together to discuss the issues that their society suffer from and it's usually conflicts between the members of their society and and we know the actors we know who how are the opposing sides what are their needs what are their frustrations and concerns they come together to to discuss those issues in detail and to find solutions unfortunately sometimes they lead nowhere they lead the only thing they lead to is the continuation of conflicts uh that that turns into violence and it could be very deadly but in in digital uh era or world peace building is is very uh has has become very challenging because of two main reasons there are other reasons definitely but two main ones I in my opinion I think are obstacles or challenge big challenges for the process are first rapid flow of information given the the technological investment we have and second is the multiplicity of actors we can in real life we can have one opposing actor for example representing the community okay it could be a political party it could be an organization it could be you know any sort of other entity in the digital world that person let's say or that entity could multiply due to the the enormous access that social media gives you to represent yourself or digital world in in in general terms through websites through you know our focus is more on on social media platforms face begins to grab twitter and other social media networks it gives you an enormous you know opportunity to represent yourself through multiple channels through multiple tools that you can amplify your voices so this this this this is also a blessing but at the same time a really big challenge when it comes to digital peace building because you then you deal with with an enormous significant source of of of people or entities that try to push to push forward their causes as well as them and might use all that that opportunity to attack the opposing sites and and what we really did in our project with the build up and was was to to firstly analyze this this this this let's say landscape who are the opposing sites what they need what are their frustrations concerns what is what are their narratives that they use and I'll go into the more detail when it comes to to the the the project I took part in and and there are also in addition to the challenges we have you know tools at our disposal that we can use in order to tackle those challenges so that's the brightness that's the good news here one of the tools that was also part of that project I really benefited from was network or actor mapping in the digital world so who are the actors who are the opposing actors what they want you know where they are located and what what they say what they write and all this kind of information is very very crucial for understanding the conflict for analyzing it or even forecasting its trends as well as most importantly to design interventions if you want to deal with this challenge if you want to build peace digitally you really need this this this this is also a tool as well as a necessity that that you need to have at your disposal in order to tackle this challenge of of of this multiplicity of actors which is on you know a huge magnitude you really need to analyze and group them together and see who are they and what they want so in the the the official term that says the network or actor mapping that's that's one of them the second one is I think part of any day to day life of a digital base builder it's the ongoing social media monitoring it is similar to the life of a journalist you know that I have that you have to be constantly aware of the situations developments whether political economic or any other sort of developments that take place around you you need to have that awareness to see what's going on sometimes it's not for the sake of your work it is a habit you it has become a I will become like a habit for you to follow all these developments so later you feel that you can that that that will be useful for you in terms of of of being aware of all the trends that are taking place so for the digital digital base builders similarly this is also a case that they have to have that high awareness of what's going on in their societies in terms of you know who are the actors the channels the media organizations that have a big presence on social media and we can affect people's perceptions on on on many aspects or whether they are a source of spreading hatred fake news misinformation hate speech and all of that that all comes from an ongoing social media monitoring for example in our project with build-up we had another tool that we used during the research project was narrative analysis okay you have monitored you have the actor map you have mapped out the actors the entities that are opposing sides to each other and you have the information what they write about how they want to lead societies or people and how they want to affect people's perceptions but what is now here that's that's called narrative analysis then you come and analyze all their talks or the perceptions what that means but sometimes it could be indirect you you won't be lucky to find someone directly threatened another group or a person or a religious a member of a religious group you can sometimes you can't find direct religious you know hatred comments it could be you know very indirect and subtle for example in this example that's on on the screen this is a poster published by a very famous iraqi television called asha phia which has a lot of followers and viewers both on social media and and on tv in one post that's about covid-19 infections they chose a picture of members of a particular sect in iraq during the religious ceremony and they wrote on it that iraqi ministry of health wants of increase in covid-19 infections nothing there is nothing wrong with this statement right what is wrong is the choice of the photo for that statement and all the people here knows that this is this photo is taken or these you know ladies are performing a religious ritual and then the negative let's say perception that it gives very indirectly that blaming this sort of religious rituals and members of that that group to be a source of spreading covid-19 however these religious ceremonies were halted during the pandemic but this is this is the kind of danger that that that you will expose during your narrative analysis by analyzing it by understanding the motive behind these kinds of state statements that are put out here we have another issue in the digital world that's for any digital please builders and they they have to tackle and they have to face it generally is the issue of hate speech misinformation disinformation my my country my region is not devoid of this i know it's it's a global issue nowadays sometimes you know societies and other that have are lucky to have correct tools at their disposal to tackle this issue unfortunately in our parts of the world this is still a growing awareness raising issue that we are dealing with but what i can say that hate speech and misinformation always said that it's federal it it can kill people i've seen it and this is also case here that people groups have been discriminated even killed based on misinformation disinformation and speech and i believe it's an important aspect for any digital please builder to deal with of course challenging misinformation is one way you know to to tackle this issue and that challenge you could be absolutely bunking misinformation disinformation on on a particular social media platforms there are a few organizations that do that interact hopefully they're growing and it could be in the comment section sometimes that you have a team or you as an individual you see a post you comment on it you try to particularly or replying to you know comments from other people other commentators to to to raise that awareness this is fake be careful and you provide evidence for example so all all sorts of these steps could be taken to tackle this or it could be one-on-one conversations sometimes in terms of in the subject of hate speech you know some people targeting another group in the comment section as we dealt with and i this was my particular focus it was under my particular focus how can i work to in order to tackle the issue of hate speech against religious minorities in iraq in the comment sections of many posts that were about you know they're the religious peace now reach will go people who are fanatics were attacking them and shaming them and spreading misinformation disinformation one way we are mighty midnight took was was was to try to speak to those kind of people and telling them this is this is this is not true people are not like that but also that conversation is i compared to a talk therapy you really can't force them to speak you really can't just say no you're wrong no it's not going to work like that you need to speak to them peacefully in order to reveal their frustrations it could be in the comment section it could be in your private direct messages or any other tool in order to let them tell you their frustrations and then you you can challenge those thoughts through that or paranoia if they have any so as part of the digital media arts from inclusive public sphere project the maps i was it was a huge and a significant steps towards digital disability busy because it equipped us as people from iraq and korea to some region with the necessary tools with the necessary process that can that can help us and identifying as well as designing interventions for this issue and in and as part of the program i was with my colleague andrew we the technical team developed system or process or tool in order to help us or the base builders to to to achieve or go which is identifying and analyzing and designing interventions and also following up on the on the interventions we made my colleague andrew will be will walk you through all the technicalities of of the of the talk and i will i'll come back later to share my experience with that gray system uh and uh it's uh it's time to present the spark andrew the floor resource perfect thanks the var so as he said my name is andrew schiao i'm a data scientist i've been for seven years also the chair of correlate netherlands which is an NGO which brings data science to other NGOs on top of that i'm also a partner at a and and machine learning um consultancy called data value people and together with them we helped build up create phoenix which is a technical platform to help local peace builders and respond to social medium so as the far said we have these problem statements that come from local peace builders there's a multitude of things that they can do interventions that they can do but there are so many people and so many places where these interventions can be placed how do we make sure that the intervention that we're placing is going to the right place and if it's the correct intervention at all for that we need to munch through data so the thing that i'm going to be talking about is the data pipeline to get all that data in how to get automatic labeling models to reduce the manual organization of the data and how to get a dashboard to get this information to local peace builders uh next slide please for the computer scientists among us this can look uh familiar it's also quite a standard structure for a data pipeline um we have data coming in for us at social media twitter facebook youtube uh event data uh we need to get it into our system which we call raw we need to organize it anonymize it do some inference on it so that it can provide value to devar to be able to do that we also need some manual labeling and once everything is processed it needs to be shown on the dashboard next slide please so yeah data sources um again if you have a cs background i'm sorry this is going to be quite shallow but we have uh apis application programmatic interfaces ways for code to contact twitter or youtube um so that we can get the data in one thing to note is facebook and instagram used to come from the crowd tangle api which meta um published however due to bad pr they are discontinuing it probably um so right now we're applying for a facebook graph api which hopefully will get us the same thing um that's where most of the data comes in um some things we aren't able to get through apis which we'll need to manually scrape in this case facebook comments we created a small plugin where um people like devar can look at a post and look at comments and while they're scrolling through comments we screen grab it and um take the comments in to our system um beyond that we also have databases uh in this case the egg led which stands for the armed conflict location and event event data which is a near real-time data of um reported political violence and protest events has locations dates actors fatalities and what type of conflict it is um what we do is we get it all in we don't change anything at all no modifications save it as is either in jason or for us park a uh next slide please now the organization this is 80 percent of my work um but we'll run through it in one slide um so we organize things because we want to be able to compare and contrast uh between different platforms we also don't want to repeat pieces of code to process each platform separately so what we do is we uh map things that are similar between platforms into one structure so uh discourse while tweets have a character limit of 280 i believe now and uh youtube comments and facebook posts don't they're all snippets of text that people are using to speak on the internet so we can map that into a bit of discourse data um same thing for authors so whether it's a facebook profile or a twitter handle or youtube account it's all a person or a thing that places this snippet of text online um important is that we map only what we need um everything else we keep but we keep separate and for this you can think of uh unofficial retweets in twitter where you put an rt at the start or certain reactions in facebook like ha ha or lol or whatever there's no real easy way to map that to everything else it might be interesting at a later point um but it doesn't map cleanly into it doesn't map cleanly into um uh one structure um here we say them as parquet files um if you want more tech uh questions ask me later uh next slide please um this to me is a funds part as a data scientist um on my classification inference so the overarching and fundamental aim for us is to build a model or a representation or theory of the system that we have an interest in uh in this case we're aiming to build representation of the dynamics and drivers of conflict within societies from how they're played out in discourse in the digital public space then you can use our model or theory um to ask it firstly what will happen with this system in the future or what's likely to happen with the system in the future and secondly what's likely to happen in the future if we were to take a certain action to affect the system so in our case we want to be able to ask something like oh please go back uh we want to ask it something like um how might we affect the system if we do initiative x now these theories or representations of systems will be something um between precise mathematical models and fully qualitative and fuzzy theories um not to do that um we need to be able to view what's going on in that system at a level that's uh higher than the micro level um to create an analogy um if you want to understand what's happening in a forest uh you don't always consider every tree as an individual you try to categorize them into species and then try to look at aggregated data or more macro level metrics uh and properties of the system so for this forest we can ask um what proportion of the forest is oak tree in the same way uh we want to be able to ask what proportion of the discourse uh in the datasets that we've uh crawled is about the economy or about gender or about sectarianism now to answer that question of what proportion of the discourse is about gender we need to find a way to efficiently um identify which of these discourse data points are about gender so this is where uh automatic classification tools and models come in um we ask the peace builders to develop a taxonomy uh that's specific to their research question ask them then to manually label a small uh sample of the scraped data um and we then use that manual dataset uh to build a general classification model um next slide please um this is what it looks like uh it's a bootstrapped google sheets because the first thing you do is you find whatever works um this has tweets in the chronological order um they can read the text and say hey is this about gender sectarianism economy um or other irrelevant uh we also give them the option to say all right there are certain keywords in there which mean that if you see this in any text it's always about sectarianism uh as well as okay there's a keyword here that usually means it's about sectarianism but not always uh next slide please once we have this um this annotated data uh we can make rule based and statistical classification models for a number of different things text relevance subject topics hate speech types of conflict and online behavior and we also ask them to make uh actor and affiliation labels uh from a pre-identified list um with this we can then run this on the rest of the dataset that um they haven't seen before haven't annotated to get classified labels for all the data so with this we can go a level deeper and ask not just how many but how are oak trees distributed within the forest are they evenly spread out uh are they in tight clusters uh analogously we can ask okay in what areas of the network are individuals um uh that are speaking about gender um is that discussion localized to certain groups of accounts is it localized to certain people um so the answer is to this question um help us build that representation uh or theory of the system and that allows us to uh gain that insight into what an effective uh initiative could be to affect that system for the better uh next slide please um so what's the value of this uh in the overarching game uh it's take all this fine-grained data uh build a higher high level representation of it um you could read every single posts um if you had infinite patients in infinite time but we don't so we need to be able to compress that data um down into something that's consumable to humans same way that if you uh want to understand the forest uh you don't analyze every single leaf now there are many open source models uh out there that compress data into something that's consumable by us and a lot of them are very good um we use this to improve our own inferences uh improve being the operative word here um as all open source models are trained for a use case that's slightly different than ours um so an open source model but might be very good at bowling down whether a particular discourse data point has a positive emotion or a negative emotion attached to it uh what it won't be good at is understanding the localized context and building that representation that we want um hate speech and the words that are used to display that hate speech uh are different in different countries and regions and it's something that I as a data scientist or programmer from the Netherlands know nothing about and it's something that um I don't have enough knowledge for to create a representation for what I can do is I can facilitate the flow of information um into the system from people like Devar people that are part of that context uh people that are steeped in that conflict or part of that community um next slide actually go go two more because we don't have enough time for this um yeah so uh let me take you through this dashboard in uh general um uh general case uh what this dashboard allows you to do is to slice and dice based on uh sentiment based on uh peace builder defined classes such as sectarianism or religion um and uh look into date and cross-reference with things such as the acled database uh it's currently made on uh AWS quick site but we're moving towards apache super set because free and open source software um but to showcase how it's used further um I'd like to hand over back to Devar great thank you very much uh uh Andrew uh on the dashboard uh that we had and uh we really uh to get advantage of was was in terms of targeting interventions in terms of for example in one case we wanted to find out uh which media organizations based in Iraq across the whole country that have uh posts uh even their own posts and the comment section that have a lot of negative sentiments and we were able to identify them easily without going through all of these uh pages and the the system gave us you know the top five uh pages that had that attracted during the certain period of time that you can choose a lot of negative sentiments and then you can narrow down later to see against tomb and what were the reactions how we're because we had different also uh the sentiments uh in the system uh for the uh the the other part of of the project was also about uh uh using the dashboard to design the intervention as I did later when I made an awareness raising uh video and also uh online campaign uh that that was wholly based on the results we got from the dashboard so in conclusion I want to say that the digital base building is if it's not uh already decayed it's an integral part in uh of of future conflict resolution methods uh I think this is this is the most important part of it that we have to sooner or later deal with it so and that's why as a local peace builder I'm 100 percent for for encouraging this this effort and improving it because that's that's the future of conflict resolution thank you very much for your time I hope it was useful and if you have any questions comments uh that we missed something in the presentation we would like to get it thank you yeah thank you very much for the interesting insights in your work and uh it was amazing how you show how technology and humans interact and build together because often technologies understood as solving problems on its own but it's always needs uh yeah intelligence how to build it and also how to use it and put it into practice it was a very interesting example for it to the audience are there already questions to the talk otherwise I I'm I would would would ask there are there are many technical aspects and and and one of them would be um you you said that you if you analyze the disc the discourse uh you normalize it so um then I would ask is this a kind of finding uh double posts and and bots for instance or and and to to to bind it back to to to the real real posters or what would be the the focus of normalizing um the twitter posts for instance um it's it's less for that I mean we also do that but that's a side effect for it um the main reason for us to to normalize things back is um we want to build a data set of this is um an interesting corpus of data for us so for a particular research question um as I said before we have lots of open source models which work very well um but for us to get something that's useful for Devar or any other local peace builder uh we need to rebuild that that corpus and retrain the model to help understand um differences between the data set that we're looking at and whatever the previous model was trained on um not too sure if that answers your question so it's mostly a data set gathering um effort yeah my question was in the direction that um some yeah you have the the danger of of a misrepresentation of discourses in in social media because if you have a a small amount of of people who are very well organized um then things can can can look much more more um relevant than they might be in in practice and and I I wondered if if this would put one one aspect of your your your data um normalization and um not really I mean it's it's always going to be inherently biased also by the the way that we do our analysis so um we don't have access to a twitter fire hose um we ask them okay what what actors are important to you what actors do you see and from there we try to expand it to um um a small subset of the public online discourse so we're always going to have that bias but um by picking and choosing actors that um are important to the particular conflict um you're able to create this narrative analysis around a conflict that they've already specified um so this uh sorry I've gone on a tangent no no it's it's it's fine I think it's getting getting clearer now um is there another question I would just open the room for questions I have a lot of but yeah um there's one question from the audience hello yeah um thank you for the interesting presentation um maybe you said it but I missed it somehow uh the the categories how many did you have how many labels did you have for many different ones and and which ones um so uh there's a difference between labels and classes so um for classes that depends on each of the the groups that we um that we serviced so Devar I think you had about 10 religion sectarianism something like that yes uh I mean the we had one about sectarianism we had one about religious hatred I think these two about that was close to each other but we have we had ethnic and others too yeah so roughly between five and 10 per group that we helped and then labeled data we had a couple of thousand tweets or posts that we used to train a model maybe let me ask another question um I still have um it's it's it's not so easy to to to make up my mind how it really is in practice there are two two things one how would you in in your local work um how would you see that that the the social media discourse is really so impacting your daily work and and how strong is this interrelation and the second one um yeah or maybe you want to ask um to answer directly okay uh I mean in terms of of my uh as a journalist you mean uh and and how social media is affecting uh you know my my profession uh I deal with every day uh it's it's it's part of my work uh you know to debunk uh to to see the you know misinformation to detect them miss or disinformation and also uh it's uh what we really focusing on right now is especially after the the research project I took part in it's just to to build an awareness about for example we have we report something on a religious uh uh uh a fast or ritual of a particular religious group inside our region we try to when we report it we try in the background information we try to provide accurate information about the event about the rituals why they do it and what's what's their significance for its people because this is the awareness we want to build through journalism to uh to our audience it in my opinion it was in the absence of this accurate information and it's in the absence of it that uh misinformation disinformation grows and people people take the chance to target them so so so to to wrap up my answer is is that I think it's it's closely uh interrelated and and and I can't really do my work without uh monitoring social media all the trends and trying my part you know without trying contributing to to to to to fix it uh you know even if it's little I think uh it's a collaboration an announced one but uh it's it's part of my work and um if there's no other questions from the audience I would I would ask um beyond your your work as a journalist um is there also a link to more to or how can I imagine the the link to civil organization groups um are there is there interest in in monitoring media and interacting then in in real life operations interactions yeah uh with regards to this question uh which is a interesting one uh in in in iraq generally and in my region because of region uh there is I I feel that there is a growing appetite uh for monitoring social media platforms because people are getting uh aware of of of the dangers uh we've we've we've already enjoyed the the the significance significance of it as the positive side of it I think people are now waking up to see it could be really dangerous in terms of misinformation in terms of disinformation in terms of hate speech and there are people you know especially young people are grouping up to to to to uh some some of them are just online initiatives others are actual civil society organizations one of them uh who is also part of their research project with us uh was tech for peace based in fact that the capital of the country uh they did an amazing job and they are still doing it in terms of uh they're also monitoring our channel as well for for any misinformation or inaccurate and they posted and they once they posted on on their platform which has a lot of um uh the audience of followers uh they uh they try to aware people of fake views of misinformation inaccurate sometimes so I believe it's a growing we are in a growing stage it's it's not fully and it's too it's not fully developed but I understand this is because of lack of awareness which is now you know uh starting to to to uh to garner momentum and to add to that um the demaps project that we were part of it wasn't just myself and of our and build up and tech for peace um we had I think 19 or 20 local peacemakers around the Middle East and North Africa uh Jordan Libya um all jumping on this opportunity to understand the digital public sphere and figure out what kinds of interventions they can place on it okay thank you and maybe a last question we as as a fifth local group for instance how we could make use of your system um it's free and open source software so great because I missed I missed the GitHub link maybe it's a GitHub link um I'll make sure it gets to you yeah fine good thank you um yeah thanks very much for the for the presentation maybe you give a warm applause to the two thank you both