Whatsoever u say, i trust u because u r my friend

Say that in the near future you will be able to post a question on social network sites. You will get many different answers (whose quality may vary). It would be nice if you could get a list of answers ranked by quality.

Problem: how to rank answers? That’s a cool problem that is disappointingly hard to solve.

Some would argue for using social networks – people close to you should be trusted and they can surely answer any type of question. For me, that’s hard to believe. That is probably because my friends have diverging interests (a few know about technology or finance, a handful of them knows about literature, and many about architecture and design – I have a weakness for creative people). And they also think differently – I’m fortunate enough to have few friends who are left-brainers, while most of them go for the “right” side. If many people would go along with my personal experience, then we could deem those solutions oversimplified at best.

So how to go about the initial problem? For a start, I would acknowlege that:
a) “X befriends Y” and “X trusts Y” are two totally different concepts. I’m overemphasizing by saying “totally”: after all, there may be a correlation between the two concepts; but it is difficult to buy into the causation arrow “I befriend you” -> “I trust you”. Therefore, that distinction may be important (if not crucial), and we usually underemphasize it “for simplicity’s sake”.
b) “X trusts Y” has little meaning. Since trust is context dependent (1, 2, 3), one needs to specify for what X trusts Y. X may trust Y for academic tips but not for real-world issues ;-). So a better way could be “X trusts Y for doing Z”, and that Z would be crucial.

Given these two points, I really like this recent paper. The authors separate social networks and webs of trust (which they call vote-on networks), and they are planning to build around context-specific webs of trust. Great work!

3 Responses to “Whatsoever u say, i trust u because u r my friend”

  1. Neal Lathia says:

    I agree with the fact that social networks may not capture trust relationships. Take the myspace example, “X is friend with Y” could mean:

    I’m sure that Obama does not trust the 196,217 people in his friend list equally. I’m sure he doesn’t know half of them; maybe he is just trying to win the politician myspace race?

    In other words, everyone participates in social networks. Bands, companies, clubs, people; the nodes in the network are no longer the same entities, and the links don’t mean the same thing. Isn’t this just resolved by making users specify their links to others (reminiscent but not as limiting as facebook’s “how do you know this person”)?

    p.s. Enough links in this comment?

  2. Tim Finin writes about a piece of software that attempts to tap the wisdom of crowds. One of the problems is how to categorize queries:
    “… the wisdom of crowds is not uniform. Since your Hoosgot query is going out to a very broad group, a narrow question on an obscure aspect of Java programming will be a head scratcher to most”

  3. [...] down in utter apathy towards the issue of trust being context-dependent – if (context=category of trust), as “rock music” is in “I trust you for [...]