A couple of days ago a friend sent me a link to The Real Life Social Network. It's a presentation by Paul Adams, the Senior User Experience Researcher at Google, touching on offline social groups, online social groups, strong and weak ties, contextual personaes and privacy. The privacy bit in particular reminded me of danah boyd's rather wonderful Privacy and Publicity in the Context of Big Data paper, echoing the idea that privacy isn't a matter of how much or how little you share, but of how much you understand (and can control) the context of sharing. So the usual problem that a status update intended for three or four people gets seen by three or four hundred and has the potential to be seen by three or four million.

The presentation illustrates the problem using the real life example of Debbie. Offline she has four distinct groups of "friends", online they're all bundled into one big bucket. Which makes it problematic when she wants to share something with a subset of people in that bucket. It's a common problem but how does it get solved? Slide 84 of the presentation says:

Allow people to create custom names for groups, and allow people to rename the group if it changes over time.

Allowing people to create groups and share only with people inside a group is the obvious way to solve the contextual sharing problem. And the Flickr friend / family split is an obvious example. But online groups have problems. Someone has to decide who's in the group and who's excluded. And if the name of the group can be changed someone has to have control over editing it. It all comes down to who controls the group. I could create an online group of school friends: Alice, Bob and Chris. But what if Alice hates Bob or Bob hates Chris or Chris hates me but is too polite to say?

Which I guess is my problem with Twitter lists. Anyone on Twitter can create a list and add anyone else to it. And give the list a descriptive title to make a claim that this group of people can be described by this label. But the people involved can't make a counter claim and have no control over whether they're on the list / in the group or not. Lots of people add the usual disclaimer to blogs and Twitter profiles that these are my views and don't reflect the views of my employer only to be added to some list bearing the employer's name by someone they may or may not know.

My other problem with Twitter lists is they make the context of sharing more opaque. Someone might follow a list you're on without directly following you and since you never get a following email you tend to forget they're there. And the same goes for retweets. But that's a side issue.

So if user defined groups don't work (and I don't think they do) what other mechanisms are there to share with a subset of people from a big bucket of 'friends'? Rather than categorise people into groups you could try categorising your relationships with people. Which is the XFN approach. But again I find XFN a bit creepy. It's easy to say you've met someone or they're a colleague or sibling or a spouse but then you hit the 'friend' word again and have to decide on friend, acquaintance or contact. Which is rarely an easy decision. It's a bit like making a new friend on Facebook and being faced by a ten drop-down choice of how you know them and where you met. It all just adds friction. And whilst I have no data about this, I'd bet that a lot of people lie. Because who wants Facebook to know you met someone in Greece, shared a house for a while, were engaged for 6 months but parted amicably when all you want to do is post on someone's wall.

I think there's a wider problem here. Both grouping friends and categorising relationships are ways to express your world view onto sets of other people. There's a lot of talk about social graphs but every time you see a social network diagram it has one person in the middle (you, the author?) with links out to some other people and possibly some links out from them to a few more. But it doesn't look like a graph. Or if it does look like a graph it looks like a very egocentric one. So to solve the problem of sharing in context maybe we need to step outside / above the graph and think less about how we connect to other people and more about how other people connect to other people. Because a group isn't defined by how I identify or label it but by the density of the interconnections within it.

Chatting with Tom about this today he pointed out that lots of social network sites use the wider graph of connections to recommend new people to connect to. So if I don't know Dave but I do know Alice and Bob and both Alice and Bob know Dave the service will recommend I connect to / follow / friend / whatever Dave. LinkedIn does this to recommend new contacts and Facebook does this to recommend new friends. And I think Twitter did do this to recommend new followees. But since new Twitter came along I can't find it anymore. (That said I can't find anything in new Twitter and it just keeps telling me I'm a bit like Tom which is reassuring but not informative.)

So this made me wonder if anyone was using the wider graph (beyond who you're linked to) to give a better sense of the context of sharing. When it struck me (somewhat late admittedly) that this is exactly what Twitter did when they changed how replies work. I have no idea why they chose to change this and whether it was about making context a little more explicit or just about managing server load but...

..in the old days when you posted a status update to Twitter anyone who followed you (and you hadn't blocked) could see it no matter what the content. Sometime ?this year? Twitter changed how this worked. If the tweet started with @alice only people that followed you and Alice would see it in their timeline. At the time lots of people (including me) complained but in retrospect it feels like a good way to make sharing contextual to a group. At least if you define a group by it's interconnectedness and not by your own definition. It changed the display logic from a line (x follows y) into a triangle (x follows y and z, and y spoke to z). Which isn't exactly a rich graph but is at least not a line.

So maybe other services already do this and make context as a product of interconnectedness more explicit but I can't think of any. And I wonder if it could be expanded further. At the moment it's only possible to separate out one person (the repliee) from any other people mentioned. If it were possible via better annotations to separate people who were the subject of a tweet from the people who the tweet was (primarily) aimed at you could restrict the context to only people mentioned and people who follow you and all the intended recipients. Would that work? Or just results in lots of #reallyfixreplies?