Skip to main content
Search form
  • 00:00


  • 00:10

    JUSTIN GINETTI: Hi, my name is Justin Ginetti.And I'm from the Internal Displacement Monitoring Centerin Geneva, Switzerland.And the issue that we work on is internal displacement,which is when people flee their homesor their place of residence.But remain within their country, theydon't cross an international border.And about 20 years ago the United Nations member states

  • 00:32

    JUSTIN GINETTI [continued]: asked us to begin reporting on this issue back to themon an annual basis.So people flee for a number of reasons.Mostly due to conflicts, also disasters, developmentprojects, and human rights violations.And our challenge is to actually collect the data.And report on it in a consistent way across different country

  • 00:53

    JUSTIN GINETTI [continued]: contexts and from one year to the other.In many cases this presents a real challenge.Because the governments in questionmay be a party to the conflict which is actually displacingthe people in the first place.So there are a lot of disincentives for themto collect this data and share it with us sothat we can do our reporting.So what we do is we bring together

  • 01:13

    JUSTIN GINETTI [continued]: lots of different kinds of data from different typesof sources.And triangulate it to infer displacement figuresthat we receive from these different sources.So for example we have some mathematical modelsthat are based on hazard impacts on homes and infrastructure.And from that we can sometimes infer how many people

  • 01:34

    JUSTIN GINETTI [continued]: have been displaced by an earthquake, or a flood,or a cyclone.Other times we might use satellite imagery analysis.And infer from the number of homesthat have been destroyed how many people were likely to havebeen displaced or the area that has been inundated by a flood.Then we also use some other non-traditional methods.So for example, Facebook has begun

  • 01:56

    JUSTIN GINETTI [continued]: sharing their data with us in the context of disasters.And from their user location data before, during,and after a hazard has occurred wecan also begin to infer not only how many people were displacedbut likely where they were displaced from.And then where they have fled to, and possibly for how long.In other contexts at a global scale,

  • 02:18

    JUSTIN GINETTI [continued]: one thing that we've just begun trying to dois something called natural language processing and machinelearning.So basically what we do is we take all of the world's news.And our algorithm reads the news and then classifies itas either relevant or irrelevant.We don't want information or hip displacements.We want it about displacements of people.

  • 02:40

    JUSTIN GINETTI [continued]: And then once it to type determinesthat this article is relevant it begins extracting featuresfrom the text.So for example it pulls out how many people or familieswere displaced.So it captures both the number, and the reporting term,and the reporting unit.It also then geospacially references the data on a map.

  • 03:01

    JUSTIN GINETTI [continued]: So we know where did this occur?Did this occur in Paris, Texas?Or did occur in Paris, France?And then what it does is then my team,as we're a team of about 10 people working on this on a dayto day basis.We analyze this data.And then this becomes the basis of what we can then share backwith governments at the end of the yearonce we've had a chance to validate and peer

  • 03:23

    JUSTIN GINETTI [continued]: review all of our figures.The other thing that we do with this informationis develop models of future displacement risk.Because in many contexts, it's notgood enough just to know how many people have alreadybeen displaced.So for example, if you're workingon climate change and climate change adaptationyou're not necessarily interested in how many peoplewere displaced last year.

  • 03:44

    JUSTIN GINETTI [continued]: But you're more interested in how many peopleare at risk of becoming displacedin the future due to the impacts of climatechange or other human processes.So sometimes we use mathematical modelsvalidated by historical data to then project into the future.And understand what may happen at future points in time

  • 04:06

    JUSTIN GINETTI [continued]: in response to climate change impacts as wellas other socioeconomic trends.There's all kinds of challenges to the data we use.In many cases we can get the same kind of a data pointfrom the same source two times in a row,at two different points in time.But there'll be a change in measurement.So that you'll have a change in a value.It may go from 100,000 to 150,000.

  • 04:28

    JUSTIN GINETTI [continued]: And it could simply be due to a change in measurementrather than an actual change in somethinghappening in the real world.And then other times with other kinds of data that we use--So for example, whether it's the Facebook data or data gleanedfrom called data records there's just quite a lotof interpretation that needs to go into that data.You know, what constitutes a movement of a person

  • 04:51

    JUSTIN GINETTI [continued]: from Facebook data?When you only have a detection of displacementfrom Facebook data how can we extrapolate from that one datapoint to an entire community?And similarly with SIM cards, how many peopleare represented by that particular SIM card?Is it one person, is it two people,

  • 05:11

    JUSTIN GINETTI [continued]: or is it a family of four?So we need you we need to addressall of these issues in as transparent and robust a manneras possible before we can ever report on anything backto our partners and to the UN member states.When I started monitoring internal displacementI had no idea how challenging it would be.I think if I had to come up with an analogy it would kind of be

  • 05:34

    JUSTIN GINETTI [continued]: like climbing a mountain in the dark.So you don't really know how big the mountain is.And you certainly don't have all the toolsto climb the mountain.In fact, you have to kind of like,invent the tools as you're climbing the mountain.And over the last couple of yearsas we've began to work with more and more differentkinds of data we have a better senseof how big is this mountain we're trying to climb.

  • 05:56

    JUSTIN GINETTI [continued]: And what is maybe the most likely route up the mountain.But it's still a pretty steep climb.And we don't really know exactly all the different waysto get there which is why we need to continue using lotsof different kinds of data and different kinds of meansof analysis of that data to come up with the best evidencethat we can possibly provide.

  • 06:18

    JUSTIN GINETTI [continued]: So this is really for students getting into this.One thing I would say to anyone who's interested in this issueis that there needs to be quite a lot more empirical evidenceand quantitative empirical evidenceon the issue of displacement and the drivers of displacement.At this moment in time there's not as much

  • 06:40

    JUSTIN GINETTI [continued]: understanding of the issue as thereneeds to be given the scale of the problem.You know, at the end of 2016 therewere estimated about 65 million people that were displaced.And that's a pretty big number of people.And given that scale of the problem our understanding of it

  • 07:02

    JUSTIN GINETTI [continued]: can bear to be significantly improved.So I really do make a call to anyone interested in this issuereally to follow it.Because there's a lot of opportunityfor young scientists.Whether they're in social science,or even in the natural sciences to contributeto our understanding of this issue.

  • 07:22

    JUSTIN GINETTI [continued]: Big data is of huge importance to the issue of monitoringinternal displacement.And I'll tell you why.On this issue, there are massive data gaps.And what we're trying to do is paint kindof a comprehensive, multidimensional, andlongitudinal picture of different series

  • 07:43

    JUSTIN GINETTI [continued]: or events of displacement.But in most contexts as a matter of fact,there are gaps in the data.The data stops being collected after a few days,maybe it's a few weeks.But certainly before the number of displaced peopleit goes back to zero.And that only tells us about the number of people displaced.It doesn't tell us anything about the conditions

  • 08:03

    JUSTIN GINETTI [continued]: in which they are displaced.So we really need to draw on a lot of different kinds of dataand analyze it to both validate the numbers.But then also to understand the characteristicsof people who are displaced.And that means drawing upon all different kinds of data peoplemight not ordinarily think of whenthey think of displacement.

  • 08:25

    JUSTIN GINETTI [continued]: So Facebook data is important, potentiallyTwitter data is important.Even where they're using their mobile phones,where they're accessing cash if they'reaccessing cash electronically.These are all things that we think about.And the kinds of data that we tryto use in our own analysis of this issueso that it can be both prevented when possible.

  • 08:48

    JUSTIN GINETTI [continued]: And responded to by governments and other actorslike humanitarian actors or development actorswhen displacement does occur.At the IDMC, at Internal Displacement Monitoring Center,the biggest challenge we face with accessing and interpretingbig data sets is actually just makingthe proper inferences from it.

  • 09:09

    JUSTIN GINETTI [continued]: Do we really understand what this data represents?Because as with any model it's garbage in, garbage out.So with big data it could be big garbage in, big garbage out.It's really on us to make sure that we're interpretingthis information correctly before we make any projections

  • 09:30

    JUSTIN GINETTI [continued]: based on our estimates based on it.But at the same time the potentialthat it represents to us is such that there'sno getting around it.And it's something that we continueto work on really rigorously.So that we can be as confident in our findings

  • 09:50

    JUSTIN GINETTI [continued]: and as transparent in our findings as possible.But it's sort of the new water that we're swimming in.And we just need to adapt to it and get comfortablewith it, basically.[MUSIC PLAYING]


Justin Ginetti, head of Data Analysis at the Internal Displacement Monitoring Center in Geneva, Switzerland, discusses the use of social media to study internal displacement of people due to conflicts, disasters, development, or human rights violations. The use of natural language processing, displacement modeling, and challenges using social media and big data are outlined.

Looks like you do not have access to this content.

Monitoring Internal Displacement Using New Sources of Big Data

Justin Ginetti, head of Data Analysis at the Internal Displacement Monitoring Center in Geneva, Switzerland, discusses the use of social media to study internal displacement of people due to conflicts, disasters, development, or human rights violations. The use of natural language processing, displacement modeling, and challenges using social media and big data are outlined.

Copy and paste the following HTML into your website