11:41 — Gilad: betaworks builds companies and nurtures them for a couple of years until they become big enough on their own. Their focus is on the transfer of media. Gilad comes from the data side and he feels like where at an inflection point. There are many interesting problems to solve.
11:42 — But why is this happening now?
1. Computation and access to it is incredibly easy and cheap
2. Analysis: We have all of these communities and researchers publishing their findings and working on these problems together publicly.
3. Access to data: As people have moved their interactions online, we have this incredible access to data and what people are talking about.
11:43 — Social networks have evolved, but there are some core concepts:
Social information flow: There are dense groups of users and main actors that are a part of the digital movement.
What’s amazing about some of these tech is that they get loosely connected to groups and empower them.
11:44 — We see these structures pop up in every network space online, and for the first time we can measure these effects and run experiments. We can understand who’s central and who is peripheral. They say that everyone on Facebook is just three people away from each other. It’s much closer that the initial six degrees of separation.
11:45 — But that doesn’t mean that information easily spreads. There are places of power within the network.
11:46 — You have to understand the structure of the network. Who is central? Who has the power? Where is the bridge?
11:47 — The smart approach is to understand their behavior within these networked audiences.
11:48 — For example: At the time, the Harlem Shake was the biggest phenomenon we had to measure. But a couple of things:
- The actual dance has been around since the 80s
- The track was posted online six months before it took off
- In early February, different people in different communities were remixing it and it was jumping from group to group
- What we can see with networked audiences is when people are getting excited about it
- When you look at the users during the initial growth of the meme, you see these dense connections where people are more closely connected to each other than the rest of the graph
- We found that some of these clusters were connected by things like the same geo, language, interests (like DJs, YouTubes, French speakers, people in Cape Town)
11:49 — Then popular accounts jumped on board, like BroBible, Jimmy Fallon, etc. Now the graph is growing as different communities are connecting with the meme
11:52 — Why is this happening? Homophily: Love of same: We like connecting to people like us.
11:53 — The other effect is slightly up to us: Algorithmic Ranking. Based on aggregate data, we can predict people who will be connected to certain groups or interests. That also helps with getting rid of irrelevant information.
11:55 — For example in Daddy blogs, you can find a community of dads split by language, #SADH (Stay At Home Dads), and different interests like, “Dad Power.”
11:56 — When you see these communities pop up, we can look at an image and determine from all of these factors, you can see great reasons why communities
theres this potential for increased polarization.
You’re position in the network directly predicts the
This is the web we’ve built.
11:57 — Networks vs Groups: It’s not a binary relationship with your user base. Homophily and user ranking helps
There are distinct positions of power and there are users who are central to conversations
You have to put influence in context
Increased polarization is a bigger issue that we have to be aware of so we can identify bridges and connectivity
11:58 — Some of this promise of data being much more accessible is that everyone’s trying to predict: But I think when you try to make predictions based on user behavior, it’s hard be cause people are irrational.
11:59 — We can do a much better job of predicting based on a community. What if we understand those group of users but weighted and scored the types of interactions their working with? Taking a community and network driven approach gives us a much better prediction.
Q: Do you think this limits us in how we grow and get immersed in diverse ideas?
A: The tools and spaces are not incentivized to serve up diverse content. What Facebook wants is to feed people things that serve their own biases. But it’s a battle for what you’re optimizing for. If you’re a business, you have to optimize for dollars.
Q: Have you looked outside of T and Fb where there isn’t as much advertising?
A: I’ve done analysis on email networks, Slack, IM chat networks, Instagram (which now has advertising, but didn’t when I was studying it) the key phenomena still exists and company’s internal data sets. We still see homophily strong and people clustering around their own interests. But every platform is very different, and once ads come in, they amplify that issue.
Q: What about when a troll network bridges across divides?
A: I used to be an optimist until I started doing this work. I’ve looked at ISIS related groups, gamergate, and others. Gamergate seems to be the most trolly. They try to attack someone on the other side, but when you look at the underlying network, the gamergate network is the one that comes out. As they attempt to reach the other part of the community, they show that they are not a part of the community at all. If you wanted to be a smart troll, build yourself up as a part of the network for a really long time, and then troll them. 🙂 But don’t quote me on that.
Q: How much work have you done looking at the details people have about themselves beyond what they posted?
Q: The bio is interesting to us. For example, if you call yourself a dad in the bio, that means something different than if someone just is a dad or a part of the dad community online. Some studies I’ve done have found that the network piece is much stronger. If we look at who they’re connected to, it’s much more telling than just their bio alone.