Skip to main content
SAGE
Search form
  • 00:01

    [MUSIC PLAYING]

  • 00:09

    CIRO CATTUTO: I am Ciro Cattuto.I am the Scientific Director of ISI Foundation.ISI Foundation is a nonprofit, private research institutionbased in Torino, Italy and in New York City, New York.My research deals with high resolution social networks.By high resolution social networks,I mean the data collected through technological means.

  • 00:32

    CIRO CATTUTO [continued]: For example, wearable sensors that allow researchersto measure interactions, in particular closerange proximity in space and the skeletonof the meters and the seconds.So I guess this is just the beginning of a new erawhere we'll have more and more instrumentationsin environments, on the bodies of people

  • 00:54

    CIRO CATTUTO [continued]: with wearable sensors.And these will allow very rich new type of signalsthat we'll be able to impact research in the social sciencesas well as in many other disciplines.And I would say that, in general, the questions we'll betackling will be the old ones.But there will be also new questionsthat can be enabled by these type of data.

  • 01:16

    CIRO CATTUTO [continued]: So it's really exciting to be speakingat this moment about this, because these are reallythe heydays.These are really the first experiences.And in my own research, I startedfrom the very simple things.So about a decade ago, by now, westarted instrumenting a small scale social gathering

  • 01:36

    CIRO CATTUTO [continued]: with wearable sensors.The sensors are coin sized sensorsthat can be embedded in badges, for example, at conferencesor in hospital settings.That can be embedded in IDs, or sewninto a pocket in a gown of a doctor in a hospital.And our goal back at the time wasto see the feasibility of tracking human mobility

  • 01:59

    CIRO CATTUTO [continued]: and proximity, or proxemics, with veryhigh coverage of a specific community in indoor settings.Normally, if you consider smartphones,some of the signals are present, available on smartphones.But location indoors is only satellite.The GPS indoors doesn't quite work.

  • 02:21

    CIRO CATTUTO [continued]: Bluetooth is still very good if youwant to do loose range proximity detection.But to detect individual conversationsclose range, face to face proximity is still a challenge.So we wanted to address this.And what we did back then was to piggyback an open hardwareproject called open beacon that created very simple radio

  • 02:43

    CIRO CATTUTO [continued]: frequency identification devices that basically sent out,on a regular channel, the identificator of the personwearing the device periodically.And then, of course, you can instrument the environmentwith antennas and triangulate the position of the personand then study the proximity network on top of this.To scale out, we decided to actually re-engineer the system

  • 03:08

    CIRO CATTUTO [continued]: and actually measure proximity directly rather than by goingthrough positioning.And so our strategy is then, it's very simple.These devices are basically small computers with radios.They emit continuously ultra low power radio packets.And these packets are emitted at such a low power

  • 03:28

    CIRO CATTUTO [continued]: that they can only be received by a PR deviceif the other person is close enough.And we are communicating on the radio frequency--on the microwave spectrum of the radio frequency.And so the signal is strongly absorbed by water--in particular, by the water in the human body.

  • 03:50

    CIRO CATTUTO [continued]: And so this means that if the devices were on the chest.it's basically shining a coronal radiationin front of the wearer.And you can use this anisotropy to detect faceto face orientation relations.So we gave this system a test driveat several social gatherings, mostly conferences.And then we started scaling up to more meaningful context

  • 04:12

    CIRO CATTUTO [continued]: for research in social sciences, incomputational social science, in what today people calleddigital epidemiology.Basically everything which is enabled by the abilityto measure the scale of the metersand the seconds, individual human encounters.Over the course of the last seven, eight years,

  • 04:35

    CIRO CATTUTO [continued]: we have found some interesting things.The first is that there are some universals thatemerge if you start measuring this data across contexts.And this was somehow our vision from the start.We wanted to create a sort of atlas of human contact, whereyou have a collection of maps of interactionnetworks in schools, hospitals, rural settings,

  • 04:56

    CIRO CATTUTO [continued]: social gatherings, museums and so on and so forthto understand what generalizes across settingsand what is specific to a giving setting.And what we found, for example, isthat a very important quantity, the probability distributionof contact durations.What is the likelihood a given interactionlasts a given amount of time?

  • 05:18

    CIRO CATTUTO [continued]: This probability distribution is actually universal.What you measure in a hospital isidentical to what you measure in an office space or in a school.And if you're a modeler, this is great news.Because basically having a universalallows you to avoid measuring this in your context.You can just plug in this known statistical factinto your favorite model and go forward.

  • 05:39

    CIRO CATTUTO [continued]: We worked a lot on tackling well-known problemsin social network analysis with this new type of data.What we found by studying high resolutionsocial network in schools is a very nuanced notionof how gender homophily works with age.In this case, it's possible to measure

  • 06:01

    CIRO CATTUTO [continued]: not only the structure of the network,but also the strength of individual tiesdefined behaviorally as the amount of timetwo individuals spend in face to face proximity.So in this case, what we found isthat boys were relatively simple ways of evolvingtheir homophily setting.

  • 06:23

    CIRO CATTUTO [continued]: And what happens is that regardlessof ties, homophily increases.So boys develop more and strongerties other boys at school.For girls, on the other hand, the picture is more nuanced.What happens is that girls focus their faceto face time, as they age, on fewer and fewer strong tiesand develop simultaneously a rich network of weak tie

  • 06:47

    CIRO CATTUTO [continued]: connections to the other gender, to boys.So this was very interesting to findand an interesting connection to the existing structureas well as prompting new questions in the area.We also studied, in collaborationwith colleagues at UC Berkeley, with a groupof females in particular, the relation

  • 07:08

    CIRO CATTUTO [continued]: between behavioral social networks measuredwith sensors and mental health.So again, this was a school setting.It was a high school.And at lunch break, we were measuring two or three daysa month, in multiple waves, the social network of interactionsof students.

  • 07:31

    CIRO CATTUTO [continued]: And at the same time, we were simultaneouslycollecting self-reported information on depressionand on self-esteem.And it was interesting to see that the structureof the social network and the relative isolationof specific nodes correlates in a meaningful way,especially for girls, with self-esteemand with the notion of-- with the possibility of depression.

  • 07:55

    CIRO CATTUTO [continued]: So this is in general a new type of toolthat allows to address some long standing questions,not necessarily having the final word as any measurement tool.There are biases to this data.There are limitations to what you collect.In particular, everything I discuss so far

  • 08:18

    CIRO CATTUTO [continued]: has a strong behavioral perspective.We are just measuring one specific behavior,which is close range spatial proximity.And our definition of tie is behaviorallydefined as spatial proximity is space,which is a very strong assumption in many settings.But it's interesting to see how this type of datacan shed new light on old questions.

  • 08:41

    CIRO CATTUTO [continued]: And they enable entirely new questions,especially when you need a high resolutionsignals, a quantitative signal on the strength of the tiebehaviorally measured.Of course, as you mentioned, this type of databecome more directly relevant whenyou think in particular of infectious disease

  • 09:04

    CIRO CATTUTO [continued]: dynamics or public health.In that case, the face to face proximity of individualsfor an airborne pathogen is a direct correlatefrom the probability of transmission.It ties to the probability of getting infected.And in this case, it is directly relevantto measure the close range proximity

  • 09:25

    CIRO CATTUTO [continued]: and mobility of humans in specific relevant environments.Schools, again, are very important.To understand disease propagation,to design intervention strategies.And in general, to reason on what we can do to fightepidemics and pandemics.And this has become our main focusin this project over the last few years.

  • 09:47

    CIRO CATTUTO [continued]: In particular, we use analytical methods, mostly comingfrom machine learning and from network science,to extract structures and patterns from these networks,and then to target these structures to reducethe burden of infection, or reducethe probability of an epidemic taking off or reaching

  • 10:10

    CIRO CATTUTO [continued]: a large number of individuals in these settings.And we found new strategies for targeting class closurein schools.For mitigating, for example, the seasonal influence.Moving forward, right now, the challengeis to bring this instrument to settings where normally you

  • 10:31

    CIRO CATTUTO [continued]: don't have this type of data.So right now, as we speak, we have ongoing one of the largestsocial measurements ever.This is in collaboration with Wellcome Trust in Kenya.And what we measure there is the high resolution social networkof students in rural and urban settings,

  • 10:52

    CIRO CATTUTO [continued]: focusing on some schools and thenthrough selected index of students to their households.So we are getting, for the first time,a longitudinal picture of what the structureof a social network of children in interactionwith the surrounding households is like.And we are going to relate this data

  • 11:12

    CIRO CATTUTO [continued]: to clinical and microbiological informationon a few specific pathogens that are relevant in that setting.In particular, pneumonia and the so-called RSV RespiratorySyncytial Virus.So this is, in general, just the beginning of it.

  • 11:33

    CIRO CATTUTO [continued]: Right now, we are looking just at close range proximity, whichis a very simple behavior on signal.But we are at the beginning of the internet of thingsrevolution, where we will have more and more connecteddevices on our bodies.Several of these devices will be able to measureclinical or physiological quantities related to stress,

  • 11:54

    CIRO CATTUTO [continued]: related to behavioral responses, to emotional status.And so going forward, we'll see richer and richer signalsthat we'll be able to qualify--both the state of an individual and the tie,the relation that this individualhas to other persons in relevant settings.The challenge of using wearable sensors to study human behavior

  • 12:18

    CIRO CATTUTO [continued]: is--it's a multifaceted challenge.On the one hand, there is an issueon being able to engage people involuntarily wearing sensors.In all of the studies we carry out,we adopt a purely opt-in policy.So people have to provide explicit consent to sharethe data, as you mentioned.

  • 12:39

    CIRO CATTUTO [continued]: This type of behavioral data can become very sensitive,especially if linked with metadata you might easily have.So we invest a lot of energy into solvingthe complexity of making people aware of these typesof measurements and onboarding them in a study.This is especially critical in developing settings, where

  • 13:01

    CIRO CATTUTO [continued]: the perception of technology might be different,where the idea of being tagged by meansof an electronic device might be controversial to begin with.So what we found in carrying out this type of researchis that we need to invest a lot of time in figuring out,on top of the engineering and the logistics, of course,

  • 13:21

    CIRO CATTUTO [continued]: the human component of the study in termsof engagement of participants.We-- especially because in many settings,we leave the participant free to switch offcompletely the device or remove it altogether.So it's critical to think about this.Another major challenge is that the technology

  • 13:41

    CIRO CATTUTO [continued]: is a moving target.So what happens is that, over the course of several years,you can seldom rely on a single device.Devices keep evolving, and you're forced, continuously,to readjust what type of physical proxiesyou need to use to assess the signals of interest.But in general, moving forward, this

  • 14:04

    CIRO CATTUTO [continued]: should become a reassurance more of the sensorsbecome available.And the third challenge is reallythe interdisciplinary nature of this type of research.Carrying out high resolution social network measurementsin a developing setting in Africa requires simultaneouslyhaving expertise in how to engage with people in those

  • 14:28

    CIRO CATTUTO [continued]: contexts, domain expertise on the targeted researchquestions-- so it might be epidemiology,it might be public health--expertise in the field studies, as well as strong expertisein analytics.And a strong expertise in computer science.There are a set of non-trivial toolsthat you have to use to clean the data.

  • 14:51

    CIRO CATTUTO [continued]: The sheer volume of the data is a challenge.I wouldn't call this a big data study in the general sense.But you are recently looking at upto a terabyte of data for a study of this kind.And so the ability of using standard simple tools thatare very common in the research, in the relevant research

  • 15:11

    CIRO CATTUTO [continued]: communities, to clean up the data is actually a challenge.And you need to have on board computer scienceexpertise, machine learning expertise, data miningexpertise to actually clean up the dataand piece out the relevant sequence.And finally, when you're dealing with such high dimensional timeresolved, individual based behavioral signals,

  • 15:36

    CIRO CATTUTO [continued]: one of the challenges for researchersis really to visualize the data to get an intuitivesense of what is going on, to developa relation between the intuition about the processand actually the process objectively recordedinto the data.And so what we found over the course of the yearsis that embedding into the team design competencies and data

  • 15:57

    CIRO CATTUTO [continued]: resolution competencies is key not just for outreachand dissemination of these insights,but really for the scientific process,for piecing hypotheses out of the-- potential hypotheses outof the data and in general for getting a sense of whatare the opportunities, challenges,

  • 16:19

    CIRO CATTUTO [continued]: limits of these type of data.But the future is bright in this respect.As we have a new generation of computational social scientistscoming out of the academia, I guessthat this gap between computer scienceand machine learning and data mining on one side

  • 16:40

    CIRO CATTUTO [continued]: and the social sciences and humanities on the otherwill be more easily bridged than in the past.And so I think that the future is bright in terms of new datasources that inform old and new questionsin the social sciences.[MUSIC PLAYING]

Abstract

Ciro Cattuto, PhD, Scientific Director of the ISI Foundation in Torino, Italy, discusses the use of wearable sensors for social network research, including how they can be used, the data they can collect, limitations of and uses for that data, and the challenges they present.

Looks like you do not have access to this content.

Ciro Cattuto Discusses Researching Social Networks Using Wearable Sensors

Ciro Cattuto, PhD, Scientific Director of the ISI Foundation in Torino, Italy, discusses the use of wearable sensors for social network research, including how they can be used, the data they can collect, limitations of and uses for that data, and the challenges they present.

Copy and paste the following HTML into your website