Profile

Cover photo
Lada Adamic
Works at University of Michigan
40,254 views
AboutPostsPhotosYouTube

Stream

Lada Adamic

Shared publicly  - 
 
Inspirational interview on unusual paths to creating amazing visualizations.
 
Recently, Research at Google sat down with Googlers +Martin Wattenberg and +Fernanda Viegas, co-leaders of the Big Picture visualization group in our Cambridge, MA office.  Before joining Google, Viégas and Wattenberg founded Flowing Media, Inc., a visualization studio focused on media and consumer-oriented projects, and led IBM's Visual Communication Lab where they created the public visualization platform Many Eyes.

Creators of compelling data visualizations, Martin and Fernanda describe their work, educational backgrounds, and their take on the intersection of computer science and graphic design. Read on to learn about their data visualization work and to get answers to some questions asked by those of you following Research at Google here.

---

Research at Google: Martin and Fernanda, you both have very different and interesting backgrounds.  Can you talk a little about your history and how you came to do what you do today?

Fernanda Viégas:  I came from a background in design and art history and never thought I would end up going into high tech. As I was finishing my undergrad studies I realized I didn’t want to be a traditional graphic designer. I became interested in the web and started thinking of ways one could visualize online communities. Keep in mind that this was a time when social networks, as we know them today, didn’t exist;  I had to explain to people that the web and email were social.  I had heard about the MIT Media Lab, and ended up going there, which was my gateway into high tech.

R@G: I’m not familiar with design or art history degrees incorporating programming in their curricula. Did you have any programming background before you became interested in this?

FV: None!  Well, to be completely honest I did try to take a programming class before going to MIT but I had a difficult time with it.  When I got to MIT I started to take programming classes in Java, which I enjoyed. That allowed me to think about how to implement visual design in computer science terms.  

R@G: Martin, you obtained your PhD in Math from Berkeley.  Can you describe how you got to where you are today?

Martin Wattenberg: I was all math all the time for first 25 years of my life.  As I was finishing my thesis, I viewed the web for the first time, and I had this experience of “this is what I want to be doing”.  At that point, if you wanted to work on the web in the media world, you went to New York, where a lot of media companies were hiring.  After graduating, I went to New York and ended up working for Dow Jones on the website SmartMoney.com.  

At that time, the idea of journalism being on the web was new and somewhat controversial, so it was a really interesting job to have.  While there, I very quickly started to realize that things like charts and graphs are different on the web -- they’re dynamic, they can change in real time.   It was then that I started learning about the computer science aspects of data visualization. 

R@G: So, like Fernanda, your exposure to programming and CS in general came later in your career?

MW: Well, I learned programming when I was young because I had an Apple II+ and I really wanted to make my own video games, so almost everything I knew about programming was focused around how to do that. In New York, I tried to learn quickly and became very interested in visualization as a subject.  Subsequently I ended up at IBM in Cambridge, MA, doing work involving email research, and that’s how I met Fernanda while she was at the Media Lab. 

R@G: You came from the other side of the spectrum from Fernanda, where the graphic design aspect was was foreign to you, but your math and computer science skills were more developed.

MW: Like Fernanda, I had to learn very quickly, and I still feel like I’m learning a lot.  That’s one of the great things...Fernanda teaches me something new every day...

FV: ...I’m definitely still learning, day by day.

R@G: Martin, you made the Map of the Market (http://goo.gl/chXT9), which was the first web-based treemap.  Just in terms of the visual design, or aesthetics, did your background in math help with that?

MW: Okay, that’s a really interesting question... Math can help and hinder. I can say that, in a sense, math has a downside in that much of it can be very esoteric… but it also has the advantage that you stop being intimidated by anything else. I think my background in math gave me the confidence that no matter how I wanted the map to look, I could do it… because it’s all numbers in the end. 

With Map of the Market, I tried a number of different things and I was able to adjust it; it felt very malleable to me. I feel like that was a big help.  I have seen that some feel limited by a particular technique – people say: “I have parallel coordinates so I have to visualize it in a certain way.”  Math allowed me to not feel constrained in that way.

FV:  When I first saw the Map of the Market, before I even met Martin, I thought it was great.  He was careful with colors and typography which made it very easy to visually comprehend the data, allowing your eyes to really take in the information in a way that is workable and useful.  I think it showed that you could take a technique, like one for making a treemap, modify it, and make it more accessible and meaningful than it had been so far.  

R@G: How does this all fit in with your work at Google?  Can you give us an overview of the kind of work you do?

MW: Our charter is to come up with new ways to display data and make it understandable to people.  We have two kinds of projects that we work on:  One is making visualizations for a number of internal systems.  For example, someone might be trying to figure out whether a complicated machine learning system makes things better or worse. This can be a very subtle problem, as there are many different ways to define better or worse. So what we might do is create a visualization that can help them easily make these decisions. 

The second kind of project is aimed at public consumption.  Sometimes, we give design advice.  For example, if Google Analytics wanted to show customers how visitors travel through their website, they’d ask: Are there certain pages where people drop off, or where they are buying something? Are there funnels where people are moving from one part of a site to another? That data is really complicated because you have so many pages and so many more ways of moving between them.  We helped define a visualization technique, using Sankey diagrams (http://goo.gl/SNha4). We made some prototypes based on those which have surfaced as the Google Analytics Visitor Flow traffic display, although I have to say we didn’t do any production engineering.  

But there are other times when our group does do the engineering along with the design work.  A good example of that is Google+ Ripples, where you can see how any public post has been re-shared and visualize the spread of information on Google+.  Some of our code is in production now that generates the central visualization around Ripples.

R@G: You deal with incredibly complex data sets. From a design perspective, how do you come up with the aesthetic for the visualization? Can you describe the process by which you two decide how much information to include, and how it will be organized in a meaningful way?

FV: There’s a lot of trial and error involved, but mostly we just laugh at the stuff we come up with!  Certain data sets can be very hard to work with...

MW: ...We argue with the data....

FV: ...There was this data set for Wikipedia, where we spent an entire summer trying to visualize what an individual’s activity on Wikipedia was like. No matter what we did to this data set it would not work; we knew there must be interesting patterns there, but the patterns we were seeing were somewhat pedestrian and not very interesting.

MW:  I remember with that project that we tried increasingly complicated things – I was going on crazy tangents and at one point Fernanda suggested a very simple idea, along the lines of just putting things in rows. I was very skeptical, thinking it couldn’t possibly be that easy. It was so simple and it suddenly showed everything.  It was like looking into a microscope and adjusting a knob and suddenly it all came into focus.

After that, for two weeks we stopped doing any kind of coding, and we would just look at all of  the different visualizations that were possible; it was like exploring new countrysides.  To me, that is the fun part – getting your scientific instrument to focus and being able to see things others cannot.

R@G: When working with these multiparameter data sets, can you describe how you deal with “noisy” parameters, i.e. data that doesn’t really convey any meaningful information?  

MW: For Google+ Ripples, one of the things we wanted to initially explore was how information was spread by examining sets of messages. One of the first things we looked at was the timeline of these messages.  And it turned out to be not very useful, because the vast majority of things in social media happen in a very compressed time scale, on the order of several minutes or less.  Eventually, we decided time was not a primary dimension. For Ripples, there is a timeline at bottom...

FV: ...But it is only supplementary, only mildly important.  We also went through a different iteration for Ripples, where we examined the structure of sharing. Basically, a post is a tree – it starts with an initial post forming a trunk, and then shares and re-shares from various branches. So the first thing we tried was designing the visualization as a tree, with names in the various trunks and branches. But very quickly, as posts were further reshared, you could no longer see people’s names. So we moved to circles, where you could more easily see the sharing patterns, the structure, in addition to being able to identify who was doing what. 

The process is very much trial and error and not being content with what you see the first time, the second time, or maybe even the tenth time.  We have to think: “As a user, what am I getting out of this?  Am I getting enough from this visualization?”

R@G: But you have to know when to step back and stop the iteration, don’t you?  From a design perspective, isn’t there the danger of over-designing a visualization?

FV: Yes, but we know when that point is! When we hit a point in a visualization where neither of us can stop playing with it, when we are no longer changing anything in the visualization. That’s when we decide our work here is done. The technique is done.  We can still fine tune things like typography and colors, but if the technique shows enough of the dynamic range of the data that you just want to play with it, that’s a good indicator.

R@G: What a great diagnostic!  “We’re playing, great, we’re done!”  Going back to an earlier comment, you mentioned the usefulness and impact that visualization can have - impact that can be applied to a great many different fields.  What influenced your decision to work at Google as opposed to other career opportunities?

FV: I think it is a couple of things. There is data at Google you cannot find anywhere else, which was very appealing to me. There is also computing power here to deal with that data that is hard to find anywhere else. 

But I think the main thing is that we were brought here, in large part, to do publicly-facing things. Visualization that affects end-users. That is where our passion lies. Historically, visualization has always been thought of as this very serious and complicated technology that experts use to look at scientific data. We truly believe visualization is something anyone should have access to and play with, because so much of our day to day lives is captured in data format.  We believe the user should have these tools, they should be able to understand what a treemap, for example, can do for them.  Regardless of the visualization technique, having this straightforward access to data is something we care about.  Being in a place where you can do publicly-facing visualizations is very important for us.

R@G: Outside of your careers, if you had an infinite amount of free time, is there any particular visualization you’d like to work on?

FV: There is one that people keep asking us to do, and that we would like to do, which would be to do wind map for the rest of the world.

R@G: We had a few questions from the Research at Google audience last week that would be great if you could answer. +Elena Micich  asked, “What can visualizations tell you about influential members of internet communities, i.e. can you collect data across information flows to find 'high-value' users who are more likely to share an idea/video/concept?”

MW: One of the subtle things about social networks is that they are much more than the simple mathematical model of nodes and edges. Everything is very contextual, and one of the great things about visualizations is that they are also contextual. When you look at the spread of different messages in communities you care about, you’ll start to notice some interesting and unique patterns, patterns you wouldn’t have thought to capture in a single metric. What I like about visualization is that it gets you away from capturing one-dimensional characteristics like influence, or virality, and instead forces you to look deeply into a community you’re interested in and maybe notice things you wouldn’t have noticed otherwise.

R@G: That’s a good segue into another audience question: “Are you doing any work on quantifying virality or scoring posts, and correlating those scores with characteristics of the post (e.g. length of text/video, general topic)?”.  Can you predict when something will go viral?

FV: That would be great! In Ripples we did find different kinds of graphs that were telling us different things about the spread.  For instance, we found what we call the “celebrity pattern”, where you have one hub, one person, usually a celebrity who posts something and then you have tons of followers reshare that right away and then it dies off. It’s one tree, it’s very broad, it happens really fast, and then it fizzles out. 

And that is distinct from, say, a New York Times article or YouTube video that goes viral, where there are many different communities who are sharing things among themselves, and who are saying different things about the content that they just shared.  We call that a “distributed sharing” pattern. Could you quantify some of that? Yes, perhaps one could model it looking just at the shapes of the graphs.  We haven’t tried to do that, but we have noticed certain Ripple “archetypes”, if you will.

R@G: As a final question, how can we get updates on new visualizations you release?

FV: A good place to get notified would be to follow this page, and the Research at Google Blog (http://googleresearch.blogspot.com/).

MW: You can also follow our personal Twitter accounts (@wattenberg @viegasf).

---
5
1
Add a comment...

Lada Adamic
owner

Discussion  - 
 
In line with this week's theme of "Cool and unusual applications", let's discuss interesting applications of SNA you've found. Several have already been mentioned in this community, e.g. https://plus.google.com/103340150936709736900/posts/eWrtLSdwr1Z and https://plus.google.com/110968293304797711777/posts/EL5WuPgFn2i
2
Add a comment...

Lada Adamic

Shared publicly  - 
 
Lada Adamic hung out with 14 people. #hangoutsonairRachel Lane, Mohammed Alhourani, Carlos Turdera, Alvaro Espinel, Sri Perangur, Saiph Savage, Rama Kumar Pasumarthi, Sinisa Rudan, Sasha Rudan, Saurav Sahay, Jose Arrieta, Shahzade Medetova, Ana María Lozano, and S.Saeed H.S.Javadi
SNA at Google Research
Lada Adamic and 14 others participated
3
Add a comment...

Lada Adamic

Shared publicly  - 
 
Ed Chi and David Huffaker will talk about SNA research @ Google.
18
1
Add a comment...

Lada Adamic

Shared publicly  - 
 
Lada Adamic hung out with 5 people. #hangoutsonairPaolo Negrini, Bart J. Buter, Franziska Keller, Алексей Натекин, and Сергей Малютин
SNA projects hangout
Lada Adamic and 5 others participated
1
Jeremy Foote's profile photoDenis Parra's profile photo
2 comments
 
Very cool - who set this up?
Add a comment...

Lada Adamic

Shared publicly  - 
 
Lada Adamic hung out with 5 people. #hangoutsonairSudhakar Shivashankar, Yonathan Randyanto, Monika Muñoz, Yu Lin, and apurba nandi
Cool and unusual applications
Lada Adamic and 5 others participated
1
Add a comment...

Lada Adamic

Shared publicly  - 
 
We'll post the link to the hangout right as it is starting. The idea is to discuss some cool applications, either ones you've seen, ones you are thinking of, or ones you have developed yourself! If you are not actively participating in the hangout, feel free to step out and follow on air (link will be on our course homepage) to let others join in.
SNA hangout: cool and unusual applications
Fri, April 19, 2013, 12:30 PM

12
1
Katharina Zweig's profile photo
 
Does anyone know whether Google shows me when the hangout starts in MY time? To me, it says 19th of April, 6:30 pm. Is that the time you announced it in your time zone?
Add a comment...

Lada Adamic
owner

Discussion  - 
 
Welcome! We'll be coordinating our hangouts through this community.
14
Krissa Swain's profile photo
 
Hi Lada. Enjoying the course :-)
Add a comment...

Lada Adamic

Shared publicly  - 
 
Lada Adamic hung out with 8 people. #hangoutsonairShihHsin Chen, Harizo Rajaona, Mirko Kämpf, Natalka Zubar, Ryan Deschamps, chaitanya khurana, S.Saeed H.S.Javadi, and soumeendra atmakur
SNA
Lada Adamic and 8 others participated
2
1
Add a comment...

Lada Adamic

Shared publicly  - 
 
Lada Adamic hung out with 4 people. #hangoutsonairhemank lamba, Shane Carson, Sasha Rudan, and Sinisa Rudan
SNAcourse hangout
Lada Adamic and 4 others participated
9
1
Ryan Deschamps's profile photoColin Drayton's profile photoCutieAngel arguson's profile photoLuis Alexander Zelada Medina's profile photo
55 comments
 
i would like to join and listen..
Add a comment...

Lada Adamic

Shared publicly  - 
 
Ingredient networks generate leftovers :) : http://www.ladamic.com/wordpress/?p=518
6
2
dieter trzaska's profile photo
 
where the way
Add a comment...
Story
Tagline
Associate Professor, School of Information + Center for the Study of Complex Systems + EECS, University of Michigan
Basic Information
Gender
Female
Work
Employment
  • University of Michigan
    Associate Professor, present
Links
Other profiles
Contributor to