 Yeah. So yeah, I would like to just say a little bit about this program to start with. Welcome to MandiFix Goa and we are Das from Goa and we aspire to build social capital through these programs and today we are collaborating with Hasgi. And this talk is about social media and because it is simultaneously revealed and celebrated for its role in social movements and public discourse, it is as well it's an enabler or a hindrance to the democratic engagement of voices. Today's speaker is Mr. Sandeep Khurana, he is a researcher on tech in the public sphere. And he will present two contemporary case studies on the role of social media in Hashtag Me Too India movement and head speech. And mostly this talk will include issues of civil society influencers, bots, misinformation, information control, surveillance, etc. and their roles in the broader social media structures. And today's moderator is Dr. Padmini Raimure. This is the founder of Design Beku, a collective that works to change and shift our approach to how we think about design and technology as processes of co-creation and participation that facilitates rather than dictates. And now I would like Padmini to start the conversation. Great. Hi. Thanks very much for that introduction. It's a great pleasure to introduce Sandeep Khurana, whose work I admire, because I think he brings excellent lens onto the way we think about social networks through our processes of social network analysis, which he will talk about. I think he kind of explicate on the technicalities as to how he undertakes this analysis. And then what it tells us about movements and about kind of phenomenon that unfolds on social networks. So Sandeep will be focusing on the Me Too movement as well as on head speech on social platforms like Facebook and Twitter. And will be walking us through how he uses his analysis to explore how the dynamics of these unfold. He's a data scientist and a researcher. He is currently a research scholar at the Indian School of Business. Before that he had, he's an ex-army officer and he completed his postgraduate degree from the Indian School of Business as well. Post that he has experience managing large projects in the IT and technology sector, both in the US and in India. So yes, thank you very much Sandeep for being here and we look forward to hearing about your work. The flow of the event is that Sandeep will talk for around 15-20 minutes about the techniques that he uses for social network analysis. And then we'll take a little break to kind of take audience questions. Audience questions, by the way, please feel free to either kind of input on Zoom or on YouTube. Because we're streaming on YouTube live and then he'll speak a little bit more specifically looking at case studies such as Me Too and head speech. And then we take another round of questions. So thanks very much Sandeep. Thanks Padmini for that generous introduction and let me start right away. Okay, so as Padmini just mentioned I'll cover the talk today in these broad sections. Firstly, I'll lay the foundation in terms of both the technique of network analysis that we use to understand the social networks and also in terms of some of the key research papers from the literature or the research background. With these two foundations being said, we'll dig into a couple of case studies as the title of the talk is on both the hindrance side as well as the enabler. So I'll touch upon both diverse dimensions, one in terms of enabler for the social movements and the other how it hinders the discourse through the case studies related to head speech. So let's start right away with the basics of the network analysis. Primarily any network is built on relationships. So it is same is true whether it is in the technology side is a network of routers or it is true for the social networks. The only difference could be that the two actors or two nodes or two persons they are connected through a tie and the nature of ties could differ the nature of actors could differ and it could be machines it could be humans but we are dealing with social networks so in our case one node to another node or one person to another person and whatever is the type of connection that will be the tie. So specific to our example or our context we are interested in the ties would would mean let's say user one is friend of user two or user one is let's say retweeting user two or liking a post of user two so each of these are interaction types or these are the relationship types and that is how we gradually build the network by adding more users by adding more relationships and gradually the network will start getting built up to a to a bigger size of many nodes or many users. Now if you notice here there is a thing there is an arrow and not just a straight line and which means that there is some directionality so if user one is retweeting user two that is not the same as user two retweeting user one so there is that direction involved in the relationship type but suppose we say user one and two are friends now in such case there is no directionality so a network or a relationship in the network can be undirected can be directed and it could also be bidirectional so for example if user one is following user two on twitter user two is following user one on twitter and that becomes a bidirectional relationship. Now another interesting part is that we need not even have say direct relationship so that is something known as a bipartite network which is the figure at the bottom half of the slide which primarily means that user one let's say is using hashtag one two three and four and user two is using hashtag one two and four now because they have three hashtags used in common these they have a certain relationship by virtue of having common hashtags so those of us who are familiar with or maybe most of us are familiar with say recommender engines where let's say Amazon suggests those who like this book will like this or those who use this page on a website they will they will be expected to like another page now those are based on these recommender engines where these kind of relationships between users can be can be seen all right so we have the users connected directly which could be directed or undirected relationship and we also have users connected through a common shared say feature or shared relationship so which means there could be not a direct relationship but something indirect but is this slide moving correctly now to the next one yes that's fine okay so the next part is that we could not just have a relationship between two users we could add more users we could add more types of relationships so for example here I have the follow relationship the read with relationship the favoriting or like relationship and we have more users and we have directional arrows flowing on both sides and that's how the network will will grow and become a big network now another interesting dimension to the network analysis is that I don't just want to build the network but I also want to use the visual features so that when I look at the network the story emerges from just the visual part of it so visualization is an important part of the social network analysis and there the tools that we use are firstly we would want to use color so if we use different colors for different nodes so for example it could show the political affiliation it could show the gender so I have the users in orange dot or blue dot same way I could have the user size depicted depicting some attributes so it could be that the number of followers is the size of the node so if you see the nodes have different sizes the nodes have different colors and I could also add more dimension for example I could use a square and a circle to indicate yet another attribute which is being modeled now it is an art and a science because if you put too many such parameters again you'll have to refer back and forth what means what and then interpret and it may become too granular so those are the tradeoff decisions as a as a data scientist or as a as an art side of the visualization but then it it gives much more depth and flexibility to use these tools to to build the kind of response to the hypothesis or to the to the research questions that you may be having so same way as I said about nodes if you see here in terms of the ties also we have those flexibilities we have the flexibility of color we have the flexibility of width so width could indicate if a and b have more number of likes or more number of retweets the the color could indicate a particular relationship type and same way the I could have a dotted line dashed line or a thick line depending on the texture of the line also I could convey certain parameters or attributes and as you see I could also put the labels so these are important tools in terms of the building blocks to to build a network so it is just this short introduction but there is there is a whole lot more which we will come to but then this is what the so is the full network visible this is what the addition of nodes subsequently one by one that will lead to so suddenly you find that from let's say we had five nodes where everything seemed reasonable what to look at and how to interpret suddenly we have say five thousand or ten thousand or it may be this is also a condensed network but it could be even more than this but now we have to kind of deal with some different tools to start reducing this network or condensing this or trying to filter it so we can remove certain nodes for that we'll have to decide the criteria because we don't want to remove any nodes which have an important say impact on or which have a significance to our research question or hypothesis so one key tool which we use here is centrality measures and there is a range of centrality measures or you would have heard of betweenness of the eigenvector centrality clustering coefficient so there are those closeness centrality those different definitions of centrality are there based on which you can remove certain nodes because you will otherwise have huge network of say even millions of nodes or lakhs of nodes which will make it difficult to do the analysis same way you can also do attribute selection you can only filter on certain attributes you can filter on the centralization centrality measures or you could filter on some attribute related to the ties and and even the x or y axis you can filter out if the visually you can take the decision now once these kind of requirements are there the computational intensity of the analysis increases so what does that mean it means that the the the number of ties grows exponentially as the number of users grow so it is straight that n n minus one by two the number of combination possible between n players the same combination the same number of permutation combination will keep increasing exponentially as the number of players increase so once the size of network is say lakhs of players and millions of ties obviously the the range the memory required the computational processing power required to process that data or even to represent them visually they all go an order of magnitude increase now when these kind of requirements are there then we have to have the tools which address this and that's why this field has picked up because not just the social networks can set in 20 years back but also the tools related to analyzing them whether it is the higher processing power whether it is the cloud to give you the give you the required space and algorithms and processing power on cloud or it could be the GPUs and the technology to enable that and all other big data stuff to do allow this so this is one set of say the building blocks or the foundation which I wanted to kind of just get everyone on board as to what we do in social network analysis the other side is the the behavioral side or the social side or the qualitative side where we get into how the social media is being used because then I will merge both these streams to to then come up to the case study and start seeing how both these allow us to to be to be kind of demystify some of these these social media phenomena so social media for social movements now is it a boon is it a bane it is a kind of evolution that we have seen when the early days of the Occupy Wall Street Arab Spring somewhere around let's say 2010 11 when the Arab Spring came and the Egypt revolution and the challenge to the government was posed now these kind of movements gave a lot of promise and that is almost a decade back and that is where it became quite quite an important subject of research for the researchers as well as researchers in both on the on the technology information systems information diffusion and those part which are kind of technical or semi-technical and also on the democracy social movements the the freedom of expression and then the underprivileged communities they all also found value in this kind of medium to do to leverage it for social movements now the way the reason they found it useful one it was the aggregation of voice because you can aggregate a large number of like-minded people with their vocal say with their voice on a specific issue it could be the the political aggregation it could be gender-based aggregation it could be LGBTQ community it could be the immigrants which otherwise may not have voice and there is a speed a spontaneity and flexibility so you have the flexibility to log in and out of social media at Meville you could participate at midnight you could participate one day not be there next day you you want to be there 24 by 7 and you want to do it from the flexibility of your home so all those flexibility that came with it there was a speed of response where you don't have to depend on some courier or messenger or so there was a there was an instantaneity there is a spontaneity and speed involved with it now along with all this convenience there was also great reach there is almost global reach so as as movement like let's say me too sure that once a certain certain cause strikes card that cause can have a reach which could be even global and you would you would get the support across the globe so so that reach also worked in favor of social movements and there were crumbling barriers barriers of the kind where geography nations or or even the barriers related to the access to privileged people or access there was no barrier there is no cost so even those who have so yes there is a barrier of having internet access but once you have it there is no further barrier so much so that even anonymity is assured and freedom of expression is assured you could go to any extent in terms of in terms of the once you cross the barrier you could you could be your own you could be defining your own persona you could be defining your own choice of people you follow choice of people you make friends with and in that sense it is democratized it is it is an even platform now the other thing is that from social movements on social media gradually media started picking stories so the media synergy is there because of which it becomes attractive because initially in the early years it was just another channel and gradually it became crowdsourcing of media news stories it became both two way relationship with between media and social media and that also gave a leverage to the to the social movements where they could they could also extend to the real world and influence people who are not on social media so that was the media synergy part and then I distinguish the way social movements are organized in two different say categories one is the movements which use social media for coordination which primarily use its reach and use its say crumbling barriers or the the convenience part and that but then it is primarily an offline movement which will bring about the change and and it is primarily the social media being used only for coordination the other kind of movements are the me too movement or the oil spill related environmental movement around the BP oil spill in deep horizon deep water horizon or these kind of movements where the the connections on social media itself is the movement so the the intensity of connections the this year there is larger presence and driver driver is on the social media itself than on the on the ground and yes it will eventually have some physical component or it may lead to a physical movement also but the early action of the movement on in Arab Spring was in the streets then got coordinated better on the social media and in the case of me too the initial action there was a big amount of interaction and activity on the social media which had its effect on the on the physical world so those were the things which work for social media but then this is gradual say shift and if we can borrow from the from the song in Silsila where we say so from there the Silsila of social media and the way we have progressed all together and seen how it has morphed into a different say say beast that we look at it today so what is the kind of issues that the researchers and the and we as social media participants deal with today those are issues of misinformation disinformation fake news I don't think the audience needs any introduction but yes say eight ten years back there were hardly any fact-checker sites we possibly did not even imagine many of us that there would be a need for for such things we we did not have terms called post-truth we did not have such kind of say say even the tools like bots and all they were not there so it was very difficult in those days but yes today it seems like the mainstream is this it seems like the fact-checkers hardly get any time and they are unable to cope up with the volume that comes their way so that is one aspect then there is another aspect of hate speech which has we'll go into more detail of this but then the two big ones which we see on on on social media today the misogyny and Islamophobia both these are like huge enough for all the prominent research centers on on either the social side or on the on the social media side they both have specialized cells dedicated to kind of address these issues of misinformation hate speech and fake news there is these tools which have been weaponized to subvert democracy whether it is during elections or subversion of democracy democratic debates earlier it used to be manipulation of some surveys or manipulation through the mainstream media or some some kind of manipulation in in the in the offline world now the the social media has also become weaponized and we'll see example of that and the process of that in the subsequent sections so that that is a grave danger and because of which again there is a lot of research interest on the democratic fallout of the social media the way it has evolved over time there is the element of propaganda primarily the extreme right wing propaganda again we'll see you did in more detail the the bots the trolls so the difference between bots and troll bots are automated twitter accounts which will post certain kind of content based on based on an algorithm or logic and and they may post they may read they may take certain actions so those are the bots and trolls are humans but they spread negativity of n number of kinds and there are a lot of there is a full vocabulary how the trolls operate or this so most of us are familiar in terms of their action but on the research side there are sub dimensions to these these because that is the depth to which this whole machinery works now this is all still yet just few dimensions all that I've covered so far there is on the platform side also there are areas of concern so there is concern around the algorithmic opacity which can have its own biases and which can have its own motives for the platform which run contradictory to the user interest or contradictory to the interest of the democracies or social movements so to give you a good example if you you are watching a certain youtube video so there is a search by zenab 250 where if you are seeing a youtube video you will be prompted through autoplay videos which are of same kind and this has led to radicalization because if someone is watching an extreme view on youtube of a particular kind youtube knows by virtue of those who watch this will also like this those kind of recommender engine algorithms they will know that this person will also like this now the effect of that is by the time your video finishes you will be fed with another video with same theme and one two three four videos once you have seen your biases get strengthened your biases get deeper so that leads to radicalization and this research and the the paper observation they all became quite defining in terms of putting some controls and also to get this issue mainstream in terms of discussion the relationship between the algorithms and the way it affects human behavior and the the fallout on the negative side of the of the human human behavior so that is again yet another completely different dimension which is a subject of interest that is the selectivism and the cause overload so again within this these two are different so selectivism is because people are kind of keyboard warriors and they are continuously clicking and tweeting almost one two three five hundred even more tweets per day so that is the level of activity on twitter they because of that they are which may be good as long as you follow it up with tangible action that tangible action could mean filing rti, fir, pil or taking out a protest march holding government accountable or or doing some action on the social side helping people or doing something to follow up on what what twitter affords as a features or to help build the social movement or to serve a certain cause but then if it doesn't translate to that it actually saps you of all the energy and deprives you of that opportunity to actually do something meaningful so that is the charge and it is true to some extent or at least with a few people where they are more keyboard warriors and selectivists rather than activists anymore so that is one thing the other is there is a cause overload and there is a there is a phenomenon known as hashtag burnout or activist burnout where people who are into these kind of social causes they have too many causes to deal with because there is a daily outrage there is a daily n number of hashtags to deal with there are daily n number of causes and you can't follow up those causes to conclusion so as a result it is just about like you are again leading to selectivism and you are again just contributing to the overall noise and overall overall say volume but then who is following it who is following it up with tangible action because of the cause overload that doesn't happen and there is a prioritization which people do mentally or or cognitively and then try to address only the the top of the mind things and there is yet another dimension of surveillance control and the representativeness in terms of are the people online representative of the entire country so in a country like India where the digital penetration literacy the the adoption of social media is itself less there is that question of representativeness of the media and there is also the state intervention that is yet another dimension of serious academic research as well as concern amongst the the social activists and the community related to the surveillance control and stifling of voice when it comes to use of social media for social causes so there is a lot on this slide and there is each is a different dimension but I'm sure we are able to appreciate that there are so many forces then at work here which cause us to give it the due importance it is almost like our whole democracy has very deep connection today with the use of social media and I have picked up very very seminal four or five papers which will kind of set the context for the next part and these will give you the background as to what is happening there so this paper is very prominent paper in information diffusion and the and the information propagation on social media and the authors Vasavi and the the senior authors in RL they both have given us very great insights through a huge amount of say a very deliberate exercise which they did here so in this paper what they did was that they took about 126,000 stories from fact checker side which included both stories which turned out to be fact or a true story those which turned out to be fake news now using these what they did further was that they kind of track these stories on twitter collecting 4.5 million tweets from 3 million different people around these 126,000 say stories which had this fact checker side and they also did this collection of stories from multiple fact checker sides and also did inter fact checker reliability check to see that there is no bias within the fact checkers and do those kind of robustness checks were done by the researchers now what did this what did this research concern just pay our paid little attention to the figure on the right and the the top most say figure there is on the x-axis we have the time which primarily shows how the tweet is progressing just as okay so we have these the tweet which is originating at the left on top which shows the twitter symbol and then there are retweets so you see the retweet symbols down the time time goes from t is equal to 0 on left to t is equal to infinity on the right and as people keep retweeting so there are different people retweeting and then different retweets getting retweeted and that is how this information propagation these people track now this has multiple dimensions to it so one dimension is depth so depth means that if it is a retweet of a retweet of a retweet now that is a depth of three so ultimate depth of a particular tweet which has got retweeted to retweet to retweet now that has a depth of three it could also mean the size which means that tweets and retweets all combined starting from an original tweet in this case if you see there are total of seven retweets and one original tweet so that gives the second graph which is the size which is the total number of people who are tweeting or retweeting this so that is the size and the breadth is if i cut across vertically that at any point when i cut across how many branches are there so in those branches term it is that i have say four branches three branches two branches and whatever are the max number of branches i try to map that out so how this is progressing i can get that also now and then there is a definition of structural virality which is a measure the authors involved which was around the burst factor that how suddenly a thing can go viral and then if there are bursts around the tweet then that will give it a high structural virality because it is going viral due to inherent nature of the content now what they found is very interesting so regardless of the the wide range of the domains they found that faults would diffuse significantly farther faster deeper and more broadly and i just explained what does the farther deeper and all these means so which is kind of insightful because it is a very large-scale study and that gives us the statistical robustness to kind of agree with the authors on this the second thing is that in terms of politics when the false news is related to politics then the effect is even more pronounced so effect is even more pronounced means that the difference between true and false stories in terms of the difference of death or difference of the breadth or difference of the size these all are much wider the next thing is the false news so the authors did very interesting thing that once they found these these kind of insights they said okay let's see what drives this so let's get into the content let's see what is in those false tweets or false stories which gives them a higher virality or higher breadth depth size and so they found that the false news is more novel more novel means this is something new something which is kind of outraging us something which we find difficult to believe something which is kind of not imaginable by us now if those kind of news come we are like our mental judgment is oh this may be something which my followers or others with whom I share it with they may want to know because this is something which nobody will know because I also don't know so because of the novelty factor they spread it more but along with the novelty they also studied the emotion dimension of the content so they they classified the all the tweets or the content related to the false news on emotional parameters of fear disgust surprise anticipation sadness joy so there are those nine factor emotional say sub dimensions along which they classified these stories and they found that false stories had preponderance of fear disgust and surprise so those were the emotions involved in in the replies to the tweets related to these false stories so I'll just pause here and then take some questions before moving on to the next set of research papers. Hi yeah thank you so much that's really fascinating and I think there's a lot of material that you've covered there but one question that I was wondering about is what are the limitations of network analysis as in you know we do know that say social platforms say Twitter for example don't allow you know kind of multiple million of APIs etc and also like how would these techniques like how do they account for bad pills like you know say bots or you know IT cells or certain parties and and also how well does this work with non-English languages so if you could just maybe say a little bit about those questions. Okay so very relevant question and I would say again the question itself has so many say dimensions and the levers around which the platforms and those who want to abuse the platform to manipulate so because of that they get a certain advantage so yes to to answer some of these issues the platform has a vested interest and they have to make trade-off decisions one side they have to be compliant with the law and the morality and the upholding democracy or those kind of universal values the other side is they have the vested interest of having more eyeballs vested interest of having their advertisers on board the vested interest of allowing more users because of that they allow anonymity and that anonymity is then misused so the platforms have that trade-off decisions they don't take it as an absolute decision which say a regulator would take or which say a fair person would take that something which is at space should be very easy to kind of then take remedial action so one is the platform's vested interest the second part is the algorithms are opaque so even the platforms at times they design it with good intent but there are unintended consequences like we know about how the stem had a bias based on gender because the the stem ads were shown to google users who were male whereas they were intended to be shown to to the to women audience so there are unintended consequences because the algorithm will maximize the the target that is given in the original design if they have to maximize eyeballs they don't care it goes to the exact segment to which you don't want it to go you want to popularize stem to women but you end up showing it where you get more clicks and you will maximize clicks so the algorithms there is a problem now same way on the bots there is there are ways to detect there was in fact a viral story where somebody used techniques to identify the bots which is that your egg dpe and then zero followers and this this and then he removed almost 1.5 lakh bots by within the compliance to twitter policy by just complaining against those and the bots could not respond back because they are eventually bots so if if he could say this this these four reasons and then file complaints and he also automated that was he also can't do that for or he or she whoever it be anonymous the hacker so it is kind of that bots are also the platforms can also do it and they can also reduce that volume but there are platform interest there are vested political interest because of which these things do not get addressed that easily now coming to the other thing which you said about language now increasingly we are using AI and it does simplify our work and I'll show the difference in the results what AI gives and what the human intervention gives definitely human intervention is much more accurate and AI is much more faster and good for volume so we have to use combination of both so the way it works is using AI classified into red orange green so something is purely let's say misogynistic and that will go into clear red something which is purely clearly not hate speech and there is no negative emotion involved it will easily go into green now there is that middle chunk on which I may want to do some human intervention as a double check and then try to qualify that and use it for research and then using that I come to a balance between both because I need to cover a sizeable chunk to be statistically relevant and also not spend disproportionate time just manually reading each of the tweets but when it comes to Indian languages use of emojis use of sarcasm use of dog whistle indirect slurs and pejoratives and and and there is an example in hate speech I'll cover this in much more detail so those are the issues which are tricky there is also people are using videos people are using images and means because of which we have to also up our game in terms of research so in in my say data interpretation I convert the images into text or I convert the images by reading text from that or by interpreting the the image by using AI to then draw some inferences based on that it is kind of becoming more difficult for both sides because they are to generate content and they are using good animation or videos and images and somewhere to decode that on our research side also we have to then use more sophisticated tools to do that so it is a cat and mouse game it is never say conclusive and there is always just a probabilistic ratio that okay I could get 60% accuracy 70% accuracy which does give a defining decision that okay directionally this is what the the statistical weight of evidence is but if you want a conclusive say 100% accuracy that is always a mirage we are chasing we have a question from a viewer Nalini Bharuthula is asking whether there's a cause-based analysis around the interplay of social media and offline so basically you know what you were saying that you know sometimes we find that one paper one paper I have to present will cover that so just hold that question the fourth paper this is the first one I discussed the fourth one will cover this okay then yeah please okay so this was all right so let's move to another dimension related to the social media research now this is research from the Oxford computational propaganda research center they have a huge say funding to study this phenomenon not just in in one medium or one country but almost across the world and across the medium and then the findings are very revealing so the researchers their Bradshaw and Howard they have this report on the global global organization of social media disinformation campaign which they initially did in 2017 there were 28 countries and then the report has more details which platforms what kind of techniques they are using are they using bots are they using private parties are they using some kind of say propaganda machinery is it influencers so what way they are using the platform which platform those details are there but the interesting thing is because of their 2017 report they found that it is not just reducing but rather going the opposite direction and increasing so they did a same report with even more detail in 2018 and guess what the number of countries went up from 20 to 48 and then in 2019 to 70 so this figure I quote is from 2019 report and the link is also there to share and what this found was that about 70 countries have varied degrees of organized social media manipulation so it if you see the dark blue colors in the in the globe global map on the on the left bottom that is what indicates the dark color indicates some kind of social media manipulation going on there so it's an increasing number of countries there are state actors influencers boards private companies all in in different countries in different measures are doing all this and then there is also the finding that they have is that there is not just okay right wing is using it there is a lot of misinformation which is part of the propaganda and that's where they call it a right wing propaganda and the other reason to call it propaganda is wherever the authoritarian regimes are there there is these say patterns are more significant or defining the patterns of information control surveillance censorship threats of violence so they do good amount of data collection in terms of what is there on social media and they have a team of researchers they will also collect for all countries they will also collect news items they will do the content analysis and then do various kind of other qualitative and and also speak to the journalists and experts to get our understanding now there are three distinct ways which they find and there are more findings for those interested who can see this and there are a lot of reports on their site related to this whole phenomena of computational propaganda which is a another term for what we have been speaking so far so the ways this this takes place is a suppression of human rights discrediting the opposition and drowning out the dissent or creating so much noise that the signal or the democratic voice is drowned so this is another research which I wanted to share and needless to say I don't think anyone is surprised that out of the 70 India is there and there is there are a lot of dimensions on which India is finding the ignominy of being included in this report so the third research I want to share is shifting from this macro to micro level now this was again an interesting and very insightful research from Soma Basu she did this research on WhatsApp groups so she became member of 140 pro BJP groups on the WhatsApp in the buildup to the to the Lok Sabha elections last year and four months she collected all the messages that came to those groups to set up the appropriate control group she joined also 80 pro congress WhatsApp groups and also different other parties which were active on social media their groups as well so there are public WhatsApp groups which you can join and there are also the groups were promoted by the parties so whatever different links and there are technical techniques through which you can collect all this data now after getting the 60,000 messages she started doing the analysis of the content and there were three distinct streams of messaging what she found but the prominent finding here is the there were about one fourth of messages which were hate speech and again as I said about the first research because it is a sizable sample size and there are these official groups there are also 60,000 messages it is robustness is checked through the control groups and the other parties content and it is it is the analysis of the content manually so there is a team of researchers who will flag the content for whether it contains or besides is much more robust that way so again this is another indication at the micro level how these these hate speech or the social media manipulation these play out now coming to the question that was asked is the hate spillover to real world this slide visible okay so is this slide visible the change slide yes okay so there was a question asked that is there is is there a research connecting the online social media abuse propaganda or hate to the real world now again this is a very seminal paper and it is such a detailed paper that and even I don't have the required statistical economic kind of background to study each of the econometric and statistical robustness checks these authors have done it is almost a 8082 page research paper this is setting a set in Germany the German right wing party so what the research has done that is of interest to us so what they did was that they tried to map the anti-refugee sentiment on facebook to the actual violent crimes against refugees in municipality level so they even went to the extent of including all the controls that we can think of so a very interesting thing which I found in this paper was that if there was an internet outage or facebook outage then it fully unders the correlation between social media and hate crimes so let's say in municipalities where people are posting high number of hate messages and suddenly there is no internet for three days due to the authorities blocking a technical outage whatever reasons then all the blockage period the hate crimes or the crimes in the physical world they did not show that correlation but in the other periods when the facebook posts are there there is that correlation so internet outages were taken the vote shares for this party in that particular municipality those were taken number of internet domains the demographics the party membership do the party have offices in that area do the party have does the party have facebook pages for that municipality dedicated pages how many posts they are putting n number of such parameters are there and that's where this paper capturing the data from 2015 16 with this amount of robust research could get published sometime in 2019 so there is a deep those who are interested should definitely see this paper but this is very widely cited and it is one of the one of the few papers because this requires deep effort to show the link between social media and actual hate crime in the real world so if you see the graph it shows that wherever there is high facebook usage the the hate crimes and the post they show a near vertical or there is a steep rise in the hate crime but wherever the facebook usage is low the graph is pretty much like very low slow so it is it is a kind of research which we find useful to the kind of subject we are discussing so now there are fundamental differences the way the three platforms which we are having the maximum membership of and which is causing the maximum concern these three platforms the way they are weaponized or the way they are being used misused or manipulated so i just touch upon a couple of points due to paucity of time but if there is a query on a specific point we can discuss that there is one fundamental differences between the closed and the the open nature of the platform so if you see twitter it is largely open you have the flexibility of keeping the your twitter account closed but it is very rare so it is it is open and that's where the different dynamics work on twitter whatsapp is largely closed so again whatsapp also has public groups but then it is largely closed and which defines then the nature of membership we are largely connected to our friends and family there is also a distinct feature of whatsapp related to its mobile phone link yes there is mobile phone link possible in facebook twitter also but in whatsapp it is near 100 percent so that gives it a different dimension related to profiling because if you have if you have seen the the the cases during the election builder between tdp trs and the the electoral srdp data leakage and the the app which was built to do the profiling so these mobile phone linked databases are obtained and then based on that segmentation done and that segmentation is used to create whatsapp groups and membership of whatsapp groups are then members of whatsapp groups are then given different messages based on that segment so it's already happening it is no longer that it is in the realm of possibility because we have seen it in 2019 election it was a subject of police case complaints political battle and it eventually died like most controversies but then there was a private party involved which built up these segmentation and profiling databases so same way facebook i don't have to tell much about the way facebook did profiling there is that campre generalitica scandal where all the details are available how this whole thing was manipulated in twitter the other thing is related to the anonymity which gives the whole dimension of hate and hate also on facebook because facebook also allows anonymous accounts easily whatsapp doesn't allow anonymous account but it is closed so if you see all three have those kind of say some feature or the other using which they are able to weaponize the platform and create some kind of hate or use it for weaponization of hate so let's quickly jump to the next section and i will just pause here if there is any quick question because there is a lot to cover and i will not be able to revisit these research papers once we get into the cases yeah hi um so i think uh the earlier question was actually also asking about the relationship between kind of social movements and mobilization online and whether there's any kind of cause analysis to demonstrate that you know some forms of mobilization on social networks work better in order to support uh offline movements so is there any kind of relationship or is it you know fairly random you know kind of what kind of social movements manage to succeed on the back of social media so again good question because as researchers uh almost every time we do research on social media uh we do give this a thought that the spillover to the real world and that is where the the rubber meets the road that that is where the the governments really take notice otherwise it may end up being noise but the the the same way like i showed the german paper and went into detail to show how this requires much more data collection robustness check unless there is a clear paper which does this mobilization does answer this hypothesis related to the mobilization question uh it is very difficult to say how it is happening yes there is a research body of research which talks of collective action and connective action primarily what that means is that in one set of movements the social media is used to coordinate and the movement actions are largely offline the other set is that the connective action where where the movement is due to the the inter relationships dependencies or interactions and then the the connections on social media and that gives it the intensity and the weight and the offline component could be just resulting in some kind of individual actions rather than mass protest or those kind of things and there could be the so in in like in case of me too i i see components of both where the movement morphs from one kind of say mobilization or connections to another kind of collective action so there is this this different ways social media is used but we i i don't know of a research paper which will clearly address the spillover of online social media social movement too and to an offline uh say activity in terms of intensity or other details of the offline activity to then do the robustness checks thank you i think we have around 15 minutes left so i'll let you finish off and we have a few more questions okay so the my first case study is around the the me to india which was largely a very heavy presence on the social media primarily it was twitter based and a lot of features of the movement they are of interest because there aren't too many social movements which leverage social media and which have a context very close to to our country and to our to our society so the features of this which generate interest one there was the presence of state and non-state actors there is the national commission of women there are those other feminists and actors but then the movement wasn't really driven by these formal or organizational bodies it was driven largely by influencers it was driven largely spontaneously and no one would have predicted that on on this date this movement will take root and then it will suddenly go viral and it will kind of dominate everything else and become become the trending topic for almost four or five weeks so this is the period in september october when tamashree datha did the revelations and and then it followed up with a lot of women empathizing with the even men empathizing with the with the cause empathizing with the with the failure of due process with the with the reason to raise a collective voice and bring about change and all those issues so it is of interest from that point of view there is also how twitter had those features and affordances because of which this could become the battleground twitter could become the battleground to give you some examples twitter had introduced just about a year back also the feature of threads which was very popularly used by by the complainants and victims to out their story twitter also has the feature of the the dm direct message and the public tweet and there were a lot of people who wanted anonymity but there was also the concern on part of the influencers that they did not want fake stories or they did not want want any fake charges which will undermine the movement so they did that important work of taking the messages through through the dm verifying and then putting them out anonymously so it could provide those features where through the mix of both these kind of features they could provide anonymity they could build trust and they could also kind of provide the reach so all this led to empowerment there is also interesting dimension which in our research we we saw the professional cross-section so there were journalists there were authors and writers there were people from the music world within music world there were people from the south indian music film industry there were people from the from the bollywood or the mumbai film industry there were people from the the professional or the corporate world who all came out in support of the movement and naming people or coming up with their with their say stories on the particular injustice that happened to them so there was a wide professional cross-section and when we see the the way these interactions evolved over time and connections evolved there were interesting insights around these dimensions and in terms of the outcomes it is not for me to comment and that is also not my area of research expertise because it is I believe it is still a movement in progress and there are there are areas to feel there is a gain and there is a change in mind search especially in urban areas corporates where there is a growing say importance given to the posh committees there is a growing importance of understanding the the nature of due process or the weaknesses of that trying to address that there may be also concerns around the defamation cases where that's where I call it a mixed back we don't know whether that is a good thing bad thing because it has a chilling effect which means the people are discouraged because of the deterrent of having to fight these battles and all so there are multiple dimensions related to the gender justice which I see this more as a prism of information diffusion social media but there are those experts like Padmini or others who may have those insights to add as to what what is that play here over time so due to the confidentiality involved I have blocked out certain names but this is the sister root network which which is kind of driving the need to movement during that period of about five weeks now something which is very obvious here is that there is heavy duty action between say about half a dozen influencers so these are the people who are primarily connecting everything to everything and retweeting people and replying to people or coordinating activities outing the stories replying to journalists or doing all sort of the heavy lifting so it is it is these kind of say results that make the social network analysis very exciting because these insights kind of really look at you they really speak for themselves through the visualization and then you can't really you don't really need to dig into the numbers or dig into the individual names or the kind of other things so this was the primary inside but then this was not not our whole universe even this slide is not the whole universe I can only pack in as much as slide can show without kind of robbing you of the of the interpretability of the slide so while that was the influencers at the top level the known names and the names on which we had good amount of detail to collect for example there we had the celebrities we had the influencers who were heavy on the activity only during the me too movement we had the people who were feminists and who had say the active organizational roles maybe chairperson of national commission of women there were victims there were complainants all those people were part of the universe that we are this is a much larger universe of people who contributed but again as you see the interactions are also that much wider so when I go from the set of about hundred or top influencers to the larger world here again the the action is quite intense during that period of of the of the movement so this is covering about 16 nine trending topics and they were in the form of various hashtags and we collected data around that and analyzed that to come up with these insights and there are more if I go to the micro level so these are the top 10 influencers so there is a spectrum of say 10,000 people there is a spectrum of 1000 or 100 and then there is only top 10 so if you see this it is of interest so the left one is on the replies who is replying to whom and the right one is on the retweets what is of interest in the replies is that there is a positive and negative sentiment to replies there is a thickness of arrows which shows that who is interacting more with each other so the arrows which are thick arrows which are by bi-directional and arrows which are pink are positive sentiment or solidarity or support a green is kind of disagreement there were areas of disagreement within the movement on certain cases where within the top 10 also they are disagreements so which is kind of interesting and insightful that there were filters and checks and balances within the top influencers to kind of not not take blind decisions on the on the cases on the right side we have the the retweets again the interesting thing here also is the high degree of clustering so top 10 almost everyone is interacting with everyone and this reflects in the centrality measure or the clustering coefficient measures these all eight and people they are very thickly interacting with each other so there are no silos and on again on the topic of silos we see some differences over time so those are like things of interest and what we also show through the colored nodes is what kind of nodes are with these people so there is a victim there is a victim spouse who is lending his way there is a journalist there is a me too influencer there is a celebrity so those kind of insights and there is also visualizations where we model these based on the profession so there is the music world there is the movie world there are journalists there are others so how they interact with each other there are insights which which are related to the to those so this is another last visualization on this use case this case study where we show I discussed about the bipartite networks where people may not have something directly in common but they may be having shared hashtags so the blue color is indication of high intensity the orange color is indication of low intensity and there is the full spectrum in between so in this case the blue color in the center indicates that there is a high degree of common hashtags between the top 20 30 people who are in the middle of the movement again these names have to kind of keep confidential but these are the same people who are there as part of the influencer network and who are doing this sharing of hashtags and and driving the movement forward so I pause here for quick question or if how many suggest I go through this also and take questions at the end it's I think we have so we have one question which is about whether there is any kind of papers or publications about the duplication of fake false news or fake news across so you know whether the same kind of fake news moves from Twitter then converges onto WhatsApp and so on and so forth and that's not only is there kind of any research or insights into that that is published okay so you are saying the same kind of fake news being detected on Twitter as well as WhatsApp yes so basically fake news like how it can travel from your platform to platform or as we have also seen say television news you know a clip kind of moving from your own one platform to another so that kind of converges so again I don't know of such a research and the reasons for that are the Twitter is much easier to obtain data but WhatsApp the groups are not that easy that's where I included that research precisely because it is one of the very few researches which came out of and there is like she had that kind of foresight as a researcher to know that during the election campaign this this will happen and then collect the data but to answer the question precisely on on both the dimensions whether the the tv channels or the media in fact I discussed this with one of the fact checker site CEOs that if somebody could track the media for their news items in terms of how much is Islamophobic how much is hate speech how much is driven against the political opposition or whatever parameters then that quantification could help drive robustness in the so someone has to physically do that which is much more difficult Twitter is also difficult but it is technically like possible once you have the the algorithms or the APIs and all that understood it is relatively easy but both WhatsApp because of the closed nature and media because of the manual effort required to do the recordings manually interpret because AI will fail there it is it is much more to deal with and there is a lot of visual audio content so it is to be interpreted I don't know of such a research but it will be useful to kind of have such a research because that will that will kind of nail it right now it is like that German research became popular because of the same reason but it is it is requiring some data collection to then come to the next question and a very quick question is that is this data also mapped regionally and geographically so the the region or geography we do as part of robustness checks because many people will not reveal the location the the Twitter features allow you that as a privacy thing so when you collect data people may say somewhere in the sky you're so they'll just put some random location so it is as a robustness checks some people who reveal and that's let's say about 30 35% in my experience about that many people will reveal it accurately so if we use that as a robustness check yes but it is not conclusive so a lot many times people will use IP address or something respect fact checkers have much more detail on on this on how accurate it is in terms of location or what work arounds they have to collect but in our case we didn't get into that part yes we did collect hashtags or the data related to hashtags which were viral in India only and then we do filter out the hashtags or which which are like say global in nature so like say me too me too alone is not defining of India only so if I'm studying me to India then I have to discount those and look at me to India and use some other combinations to filter out but otherwise location is again not truthfully revealed on social media to kind of be conclusive on on the large scale thank you um yes please go so the next one is on hate speech again it is a huge topic and even I was surprised on the types of hate speech and the amount of flagging of hate speech into different types and even within this like on sexism or gender this thing there are again another five sub types or on how the misogyny is played out in in five different subcategories and for clickbait there are other categories on boards or this so there are n number of the n number of such say classes or bucketing and there are agencies paid unpaid research algorithms which do this kind of flagging and which is what we use for research starting from very simple word search to getting into AI techniques to get more traction but again as I said earlier also it is a cat and mouse game and it is a like mirage which we keep chasing so it is there are reasons for that so I will come to some more insights as we proceed so there is there is a larger say problem of hate speech which is at a strategic level it is about the narrative setting whether it is in elections whether it is in day to day the way weaponization of platforms has happened is to set the narrative which could be to distract from a real issue and set the minds on some dummy issue or it could be to set the narrative in the direction which the manipulator wants and those are the kind of things which lead to lead to the abuse or computational propaganda how it happens is that there is that innocuous looking dog whistle which means there is an indirect reference to something which somebody can disown and which is kind of under the radar but then it sends the signal to the dog or the way dog whistle works is the whistle is understood by the dog which will bite the person and if you come and catch me that why did you whistle you can't tell me that I whistle the dog to bite you so I whistle the dog in a coded message the dog came and bit you because I and dog have an understanding but you can't catch me legally because you can't say the whistle means this so that is how the manipulators whether it is Trump whether it is the other leaders here whether it is the social media influencer they will not be direct and they will use dog whistle now this dog whistle gets amplified using noise in an eco chamber where this amplification detailing and further details will be added to this dog whistle to make it more obvious that leads to a velocity and volume when velocity and volume are the parameters on which twitter adds it to the trends the reason something trends gives it even more amplification because once something is on trending topic you and I will go and click and watch it which gives more eyeballs we may react reply so because of that there is even more amplification increasingly as I said media spillover there is two ways spillover from social media to media and media back to social media because there is something which is trending on social media for example if me to India Tanushree Dutta this media may not be interested but because the influencers made it count by trending it on social media the spillover on media is there they were they were forced to cover that story so same way there are other many episodes where because something is very prominent on social media it will get covered by media because of that media spillover there are other social media spillover because media covers it media starts getting retreat so in my networks I find a lot of media embedded in specific networks because the media covers only a specific message so because of that they become entrenched in certain networks and that becomes a vicious cycle and this goes on and on and that's how the the whole thing becomes from a simple harmless tweet to something which is a weaponized platform so this is to scare you that this is what happens so what is this this is a series of hate hashtags which we collected all the tweets and players over a period of say about three months and it includes about 80 hate speech hashtags we filtered the other hashtags out and we got the top 80 hate speech hashtags and then we took all the top accounts by the number of retweets they got and got close to about 75,000 tweets which had these 80 different hate speech hashtags so this is where social media throws this at you now with this network I don't know how to figure out so I have to use some of the techniques to come to something more meaningful so as I start narrowing down to the key players using some of the techniques I mentioned in the beginning I come to the top 30 40 people now this is the set of top 50 people and what you see in red is the the leader so this is the anatomy of the organized hate where one set of people how they behave that is the set of people who are the leader in this hate speech they will give the the dog whistle or something very prominent which due to the size of this slide may not be visible but now when I reveal it to you will notice every arrow is inward pointing to this account which means everyone else even though they are in top 50 but they are retweeting this account so this dog whistle is important enough this has been registered by almost half of the top 50 people to the extent that they directly retweet this or directly act on this there is no outward arrow because the leader that way is smart enough to not say it openly or not get involved into this openly this is the first set of the structured organized which is the leaders the second set is the influencer so they are the ones who will spin it they are the ones who will amplify it they are the ones who will add context to it they are the ones who will kind of add man's images or get the message to be what the leader wanted it to there could be offline coordination there are fact checker sites who have revealed how the offline coordination happens to coordinate these kind of things between leaders and influencers so that is we don't know as researchers but we know what we see so what we see is that the influencers have very interesting patterns so what you see now in red color is one influencer who is one of the top most influencers and almost connected to everyone so almost connected to two-thirds of the network and what this influencer is doing is half the people the influencer is retweeting and half the people are retweeting the influencer so it is kind of very important connect because it has amplification because it has the ability to spin they are able to spin it they are able to generate content and also they are able to amplify content which means they have good amount of fan base and they have good amount of creativity to do both so this is where they have thick arrows with others they have they have a high number of followers they may not be obscenely high followers in millions but they have significantly high number of followers which is where they have their key number of they have key position in the top 50 so this is the influencer part the third one is again very interesting to note now this is the amplifier amplifier is also embedded in the top 50 but the unique thing you notice here is not a single incoming arrow almost every arrow or each arrow is going outward so what does it be amplifier to amplifier is only reflecting everything only retweeting everything has a good number of followers can generate heavy volume because if you keep retweeting a lot you generate a lot of say say interaction or activity and you gather a lot of volume but no originality and how you get centrality you get centrality because of the phenomenon known as reciprocity if you retweet somebody they will also retweet you they will thank you for that and things like that so because of reciprocity they get their amplifier gets their position in the network or power in the network or the reason to spend all day because they get to this top layer purely because of amplification without getting into any generation of content and then the fourth category which is what I call the useful idiots so these are the fringe wannabes who are only retweeting or who are just sometimes they generate odd tweets which the leaders or the influencers they want to kind of retweet to keep them in good humor keep them as fans but these are all the wide network who are the fringe wannabes who also someday want to graduate to influencers or who have their own reasons to identify with ideology or so that is the so that is the anatomy or the structural machinery of hate which includes these leader influencer amplifier and the useful idiots now how it plays out the dog whistle at the center and that reveals that generates a lot of activity because that dog whistle at the center leads to a lot of anger and denial and hate and even some people will report it so I just picked up one tweet from today and just reverted some of this content on the right and this is what is being discussed on twitter today since morning that someone says I have liberty and right to do so and so I'll buy whatever I have freedom it has nothing to do with religion now this is a dog whistle it has to do with religion which when you get to the comments you will get to know but the dog whistle from the leader the leader gets away nothing to do with religion and I'll do what I want I have freedom I have this then somebody in the in the amplifiers and in the influencer they start adding to this original one on the right when you see the second tweet so where is this calling for social boycott of so and so so and so just said this and not this this tweet doesn't mean this where does it say this and the influencer will also answer why are you not naming no I don't want to be reported to twitter so the dog whistle is being given by the person who knows that if they go vocal then they'll get reported so the people are asking why don't you directly name no no I won't name directly because of this so then somebody says no this is the next page we can understand you have been disingenuous blah blah this I think everyone is seeing this daily and this is the anatomy of hate and this is the anatomy of hate at each individual viral tweet level or the hashtag level of this so this is where when I collected this data from today's tweet itself within last say 10 11 hours this I collected during the day I analyzed so there is these n number of replies which so I got 750 replies 10000 retweet is a huge kind of popular viral tweet which is a hate tweet but it is a dog whistle you can't report it you can't do anything and I took a sample of 4000 by about midday and then analyze them so there is a lot of responses which you see in the middle I've just reverted and compressed them so there is I agree 100 percent agree it is that which I can use AI to kind of filter these and which is what I built the algorithm to give me how many of these tweets contain religion and how many contain agreement with the original person how many talk of othering us versus them kind of thing and how many people oppose it that no this is hate speech and we'll not agree and we don't agree to this kind of thing so these figures 2029 they indicate if I do it manually or if I do it in a algorithm way the manual thing is more accurate by about 9 10 percent at this level I just took a random sample of say 100 and then try to see how much of the accuracy is built in and the reason for that is also obvious because people will use so many indirect indirect references to religion or hate that it is very difficult to be precise so I'll just stop here I had to rush towards the end in the interest of time but I'm open to questions and that is all that I had to share for today. Thank you so much Sandeep lots of compliments over on YouTube in the comments but one really important question I think and a great question to kind of take us kind of forward based on what you've been discussing is that you know what is it that we can do about you know opacity and you know as you said computational propaganda etc like what can we do in terms of shifting policy for example is there any way to push platforms towards more regulation towards more accountability kind of what are there ways in which we can use the kind of evidentiary work that you've been doing to kind of leverage that in order to maybe changes or push for changes in regulation? So to be fair there are already a lot of policy advocacy groups there are a lot of presentations made by this group we have a center for internet freedom there are other say research agencies social NGOs and other bodies which are dedicated to particular causes related to privacy surveillance and they are engaging with the government but as I covered in the research also and there are 70 countries with say that kind of machinery and there are also countries with authoritarian regimes who are not interested to engage and on the part of platforms if the government and the civil society they both come on same side so on certain issues the government is also a victim of hate speech so like in the pandemic even they became victim of misinformation and disinformation no don't wear mask or wear mask or this so that and or targeting the community those things then hurt the government also but the moment this alignment happens right now if the government or the regulatory bodies they themselves show disinterest to engage with the policy advocacy groups it becomes difficult because then you have to overcome this additional resistance and so researchers contribute by arming these advocacy groups because then they have the backing of say the statistical robustness and they are not talking just based on hearsay or observation they are robust in their claims and then someone has to engage I mean as we said about spillover and that engagement all stakeholders and so we have seen like for example I can recall from the pre-election monitoring of social media before 2019 social media there were policy representation on how it can be controlled and how it has to be reined in but eventually election commission pushed it so close that the only thing they decided was to have self-regulation or self-regulation is an oxymoron it it doesn't happen with the platform regulating themselves it did not really yield much results and again I have enough evidence from the from the 2019 election also to suggest that there was no regulation there was no control on hate speech during the period many people have asked whether your slides will be available after this and which publications talk about this research that you've done whether that's so both the research papers are under publication one or under research one is already submitted but we have to wait for the reviewers comments and then so that's where most of the content I have to block out the the identity is named or even otherwise I have to block out because of the confidentiality but right now so I come from the scholarly side of the information science where we deal with the information diffusion propagation and the research of Aral and Vasovi that I suggested shared in the beginning where the research the information goes deeper wider and that part so I don't have social this thing but yes unless I take a context so only to answer the question so far both are not published one is under publication the other is under finalization the the slides will be available but with some bit of pruncation and the you're most welcome to connect with me in case you have certain joint research proposals or you come from a discipline which has a synergy with the kind of work I do and the tools I use so we can we can we can discuss that okay thank you so much I think it's been a fascinating conversation and there's a lot of interest and engagement on YouTube so thank you so much we are trying Sandeep if there are more questions we will pass them up to you and hopefully we can get to them and answer them later so yes thank you everyone for oh yes yes just worth mentioning that Sandeep himself is not on social media probably very that was that was that was my last slide where if you recall that was the slide from Cal Newport he has given a TED talk saying quit social media and I found it very kind of convincing so that's why I share this link so the title is quit social media and you you listen to this talk and it is only recently that I quit it allows me to focus it allows me to focus on on on deep research so if I have to do good research the problem is social media is very distracting and it consumes disproportionate time so I do have anonymous accounts to just kind of keep track because of the research interest but I am not a social media person and there are many researchers in my area who do that who are including Cal Newport himself he's a tech evangelist who has written a very good book on deep work so yes I think it helps you to be more objective about it yes even that is there because I don't have to take sides but I report what I observe okay thank you so much everyone for tuning in if you enjoyed today's talk next week there's a conversation again hosted by Das and Hasgeek by Vivej Joshi on oh no in India no no sorry no sorry it's on the 25th sorry it's on the 25th of May and the if you go to hasgeek.com slash Das at THUS you'll be able to find out more information about that though and then in the week sorry is another conversation between Pratik Senha again and Denny George Pratik of course is the co-founder of Palsh News and Denny is the co-founder of Tattler which is a survey technologies platform that aggregates fake news and so they will be a conversation with Pritthi Raju Khurji and that's Monday 18th May from 5.30 onwards a different time so so yes please join us for those conversations and thank you very much thank you it was very nice thank you