Professor Karrie Karahalios is a current Berkman fellow, joining us from UIUC where she teaches computer science. Her talk at Berkman today is titled “Text and Tie Strength”, and begins with a reminder that “What attracts people most is other people”, a quote from sociologist William Whyte. Whyte pointed out that people flock to spaces where they can hang out – places with seats. As we design online spaces, we need to consider building spaces with seats.
Karahalios worked with Judith Donath at the Media Lab, building visualizations of ties through text. One of her early projects colored messages in Usenet group soc.culture.greece, identifying the sentiments of each and painting the portrait of a very irate group. Other work includes “social mirrors”, ways to visualize conversations to teach turn-taking to children with Asperger’s. Some visualizations enable votes in online systems to have a persistent identity, even if people’s motions didn’t succeed, as people want to see their vote, even if they don’t get their way.
In studying the nature of ties, it’s important to understand Karahalios’s personal background. She grew up in a small village in Greece, a town with 1000 people when she was living there, and fewer than 300 today. Her father would call from the US at noon every Sunday, and she’d take the call from the local tavern. Ten to fifteen people would show up to listen to her half of the conversation, perhaps because television didn’t go on the air until 5pm. Once her grandfather got a phone, people would show up there as well to listen to the entertainment.
It’s possible that usage of social media might be very different between rural and urban communities. She points out that prior to 1920, there were more phones in rural than urban America, and that people used them very differently. As in her small Greek village, the telephone was entertainment, with farmers hanging out on party lines to hear what was happening locally. (She shows a 1920s (?) ad that shows “how the telephone makes farm life more enjoyable”, ensuring that “the old time isolation is a thing of the past”.)
She and grad student Eric Gilbert sampled 3000 MySpace users and segmented them into rural and urban groups, and studied their posts and interpersonal messages. They sampled to create two roughly equal sets (while only 24% of the US population lives in rural areas, defining rural as “living in towns less than 2,500 not connected to a metro area”.) They discovered quickly that rural people logged on more often than urban users, and they set out to test five hypotheses:
The rural population will have:
h1 – fewer friends and comments
h2 – more women
h3 – more private profiles
h4 – closer (geographically) friends
h5 – preference for strong over weak ties
They were able to find strong evidence for hypotheses 1-4, and especially saw evidence that women were the keepers of privacy within networks. But they couldn’t prove or disprove h5, because it’s not clear how to characterize strong or weak ties. While the idea of strong and weak ties is a critical one in social networks – largely due to Mark Granovetter’s work on the idea and its implications – there’s been little work done moving beyond a binary definition of strong and weak ties. She’s curious to explore the dimensions that represent strong ties – time, emotional intensity, the “reciprocity of emotional services” – and to see whether there’s evidence for more of a continuum between strong and weak ties in online environments.
It’s challenging to study these ties in online spaces in part because the tools aren’t very subtle in how they characterise a friendship. Facebook, for instance, has a binary definition of friends – people are either friends or strangers, but your spouse is given equivalent weight to a guy you met at a conference once. Karahalios shows a visualization of her Facebook network – it’s got large, interconnected clusters and a single connected triad on the outskirts of the map – her only three Greek relatives who have computers!
To add some richness to Facebook’s binary nature of relationships, she and Gilbert interviewed 35 university staff and students, and had them characterize 2,184 friendships in terms of questions like:
– how strong is your relationship?
– would you ask this person to borrow $100
– would this person help you find a job
– if you left Facebook, would you bring this person with you (important to think about when we consider Friendster’s rapid evalporation)
The goal was to correlate people’s reported strength of ties, as determined by these interview questions, with network indiators visible within Facebook: the numbers of wall posts, the numbers of wall words exchanged, the number of inbox messages, days since last communication, as well as work around sentiment analysis of message content. Their analysis determined that a working model considers these factors (in order of increasing importance) – structure, emotional support, services, social distance, duration, intensity, intimacy – and that good predictions can be offered from 15 variables.
One of those most important factors is the “days since first communication”. In social networks like Facebook, we evidently tend to add our most important contacts first, so this factor is highly correlative. But other indicators are more complicated. It turns out that many people maintain strong ties to people they’ve broken up with. She points out that “hate can be a strong tie” and that some of the most interesting discoveries so far are novel behaviors that break the model.
Given an algorithm that works pretty well for differentiating strong and weak ties, Karahalios suggests that we might restructure our tools around these capabilities. We might sort our photos so we released some to our strong ties and others to our weak ones. We could have alternate friend feeds, prioritizing messages from our more strongly-tied friends.
Armed with a working model, Karahalios and Gilbert wondered if they could determine strong and weak ties in a different community: Twitter. Despite some superficial similarities, Twitter’s a very different beast from Facebook. There’s no assumption of reciprocity, a lower barrier to entry, and a population that’s smaller, though still huge. Their new experiment is called “We Meddle“, and based on the quantitative indicators from the Facebook experiments and sentiment analysis of the texts, the system offers clusters of your Twitter friends, including an inner circle (strong ties) and outer circle, as well as other possible clusterings of people in your Twitterverse.
The system is open for you to experiment with – you’re encouraged to “tune” your groups, dropping inappropriate members, and then use a Twitter client built into the system. The client prioritizes messages for you based on group membership – messages from your in-group are larger and more visible than those from your out-group. The userbase is pretty small right now – about 600 users – and they’re hoping for a couple thousand to test the tool thoroughly. (They’re likely to contact users for qualitative analysis after they’ve tested the tool.)
I signed up immediately, and was very impressed with two of the clusters the tool produced. One contained friends from the Global Voices community and also from other international projects I’m involved with – I termed it “Cosmopolitans” – and the other represented prominent tech pundits and alpha geeks who I follow – “Digerati”. The in and outer circles were somewhat harder to evaluate. The truth is, many of the people most emotionally important to me aren’t on Twitter, so evaluating a group in terms of completeness is pretty hard. I looked at the inner circle list and pruned folks who I follow, but don’t feel personally close to, and got a list that feels slightly more like an inner circle… but I also feel like I got a good sense for just how tricky this line between strong and weak ties actually is.
David Weinberger has an excellent account of the conversation that followed the formal presentation – I’ll link to him, rather than attempt to reconstruct a conversation hours after the fact.