Karrie Karahalios is giving a Berkman lunchtime talk entitled “Text and Tie Strength.” Karrie is a Berkman Fellow from the Univ. of Illinois.

“What attracts people most is other people,” said William H. Whyte. People want to sit and talk. But what is the equivalent of a seat in a virtual space. Her group studies visualizations of group interaction. She shows examples, including how voting changes interactions, and how conversations cluster around topics.

Karrie grew up in a small town in Greece. Every Sunday her father would call from America. The call would come to the one phone, which was in a tavern. The people around expected the call and would participate. People use communication media differently in rural and urban areas, she says. E.g., rural communities used to like party lines; it was like a sub-net. Urban folks didn’t and the telephone companies moved to individual lines.

Her group sampled communication usage in rural and urban communities. They had five hypotheses. Rural people would have networks with:

fewer friends and comments

more women

more private profiles

friends are closer geographically

preference for strong ties over weak ties

The results showed that all but the last were true. In part this is because it’s hard to quantify strong and weak ties.

She suggests: Maybe there are two MySpaces: Rural and urban.

To understand weak vs. strong ties, her group explored Facebook. It’s quite binary: someone is either a friend or a stranger. The dog breeder who you met once has the same presence in your FB network as your husband.

Tie strength was invented by Mark Granovetter in the 1970s, in the book Getting a Job. Karrie says: “Strong ties are the people you really trust.” They help people through difficult times. And if a strong tie is depressed, you might get depressed also. “Weak ties are merely acquaintances.” Granovetter pointed out the utility of weak ties in “The Strength of Weak Ties.”

Her group looked at FB and wondered how to map FB parameters to tie strength. They set up a set of questions with continua, e.g., “How strong is your relationship with this person.” They assessed 2,184 friendships, from 35 university students and staff, along 70 parameters. E.g., friend-initiated wall posts, wall words exchanged, friend’s status updates, inbox intimacy words, together in photos, age differences, political differences, mutual friends, groups in common, links exchange by wall, applications in common, positive and negative emotional words, and days since first communication.

Her model had seven elements of tie strength: structure, emotional support, services, social distance, duration, intensity, intimacy. Her findings showed the relative importance of each of these (which I’ve listed in order, from least to most). The most predictive FB element was days since first communication. This may be because the first people you first connect with are the ones you are most tied to, although you may then not use FB for much communication with that person.

Karrie finds it quite interesting when her model doesn’t work. E.g., “This friend is an old ex” who was friended when they first began. She says that strong ties can be love or hate, although we tend to assume strong ties are positive; her model doesn’t include negative strong ties. Also, there are times when someone else’s account is used as a proxy for two others to communicate, e.g., neighbors who are feuding and only communicate through a three year old child’s FB account; Karrie’s model does not account for that.

How might this applied? Suppose you could organize your photos so your strong ties saw one set and your weak ties saw another? Trying to do this by hand is a nightmare.

They did this work in 2008. Then they wondered whether it applied to Twitter. The created wemeddle.com where you can see the people you follow on Twitter. It clusters them by the strength time. Photo colorization and size indicate strength. Karrie says she’s been surprised to find that she’s more interested in what her weak ties tweet.

Her group is studying the quantifiable data (e.g., server logs) but will also interview users.

Q: Your model users linear regression on Facebook?
A: Yes.

Q: How about a geo-map visualization?
Ethanz: Someone recently did this sort of thing for Facebook. There are areas with cross-national friendships and some without many.

Karrie wonders if there is a single model for strong and weak ties that applies to all social media.

Judith Donath adds that following links is a strong signal of a strong tie, which is information that the WeMeddle client could start tracking.

Karrie: In FB, we took advantage of reciprocity as an indicator of tie strength, but reciprocity for Twitter doesn’t work.

Ethanz: Strong vs. weak is so murky. WeMeddle is a very nice provocation. Also: LiveJournal gives you valence, as opposed to FB that only lets you friend or not friend. At FB, every relationship is symmetrical. Twitter is more like celebrity: once you have over a few thousand people, you’re broadcasting. It’d be interesting to look at tech that enables a strong-weak tie continuum.

Karrie: There’s lots of lit on info flow, but not on how the strength of ties influences how you send info out across the network.

Judith: We’re all fooled by the asymmetry of Twitter, which makes for a bizarre set of ties. WeMeddle and your model might help us make sense of it.

Donnie Dong: Could WeMeddle combine FB and Twitter?
A: Yes, it’s possible. People would have to put in both logins.
Donnie: Twitter and FB are blocked in China. It’d be interesting to look at people who communities inside or outside the Wall.

Q: Have you looked at email? And have you looked at the connections among people twittering about disasters?
A: We haven’t looked at either of those things. Disaster relief would be fascinating to look at.
Ethan: There’s a report that during the Iranian uprising, there were only 60 Iranians tweeting, and the rest were Americans retweeting.

Q: Can you track how people gain trust, to move from outer circle to inner circle?
A: The trust problem is really hard. Trust with text — the literature hasn’t been very complete. With 140 chars you don’t get a lot of queues. On the other hand, looking at reputation systems might get you somewhere. I wish I had a better answer for you…

Q: [me] Doesn’t this suggest that strong and weak ties is too much of a polarity for the Web?
A: It suggests that trust doesn’t map to strong and weak. There may be several types of strong and weak ties.

Judith: Granovetter was really interested in homogeneous and heterogeneous ties.

Ethanz: Maybe look at John Kelly’s work on how blogs link to third parties. Look to see what everyone links to. You’ve got all the data, but if you grab the links people are linking to, you can imagine another way of clustering people.

Karrie: There’s lots of work recently about how information gets dispersed. E.g., to spread info quickly it’s better to have a network of people who believe things easily than having one large influencer. Also, it’d be very interesting to map people by complementarity, not just similarity. Overall, I’m really interested in how these ties evolve over time.

Wendy Seltzer: You could look at contextual work (a la Nissanbaum). E.g., do people name their groups at WeMeddle the same way, and how do people move people in and out of groups.
A: There’s a set of privacy questions. Suppose people publish the list of their inner circle. Why aren’t I on it? I don’t want to destroy any relationships with this work. What attracts people the most is other people.