Re: collective intelligence - bayes theorem help

["Kyle R. Burton" <kyle.burton@gmail.com>]

2 things: first, can we _please_ have more of this kind of discussion
on the list, this is an area where I have little to no background and
don't understand and I very much value this kind of discussion.

Second, Steve - based on what you learn, can you do a presentation /
talk about using conditional probability for social networking
recommendations (at least what you've posted sounds like this).  I'm
sure our group will be forgiving about your level of experience - what
you'll have learned will be new and valuable to most of us (I'm
assuming).

Thanks!

Kyle

On Thu, Jul 31, 2008 at 2:03 PM, Jonathan Tran <jonnytran@gmail.com> wrote:
>
> On Wed, Jul 30, 2008 at 9:24 PM, Steve Eichert <steve.eichert@gmail.com> wrote:
>> A = Person X will identify Person Y
>> B = Person X is in the Philly Lambda user group
>>
>> However, in order to take this approach I believe I would need to know the
>> probability that person X will identify person Y, which is what I'm trying
>> to figure out.
>
> I'm no probability expert either, but have you tried solving the
> conditional probability formula for P(A)?  As in...
>
> P(A|B) = P(B|A)*P(A) / P(B)
>
> P(A) = P(A|B)*P(B) / P(B|A)
>
> ASCII text may be a little misleading.  Your events are actually
> parameterized over X and Y.  Normally this would be written B_X (TeX),
> as in B with subscript X, to represent the event that Person X is in
> the Philly Lambda user group.
>
> The reason I bring this up is because to figure out P(A|B), we would
> take the number of people in Philly Lambda who identify person Y, and
> divide by the number of people in Philly Lambda.  And this would be
> different for each person Y.
>
> The weird thing, which I think may be the cause of your confusion, is
> that for some people, we don't know who they identify.  We can't
> really compute P(A|B) as I described.  So do we include them in the
> total number of people in Philly Lambda?  ... You see what I mean?
> Because they are unknown for who they identify, we can't really say.
>
> A simplification might be to exclude the people who did not respond
> from the dataset.  Use that dataset to compute the probabilities.
> Then predict the ones who didn't respond from that.  I think this
> makes sense because it's like spam filtering.  You use all the emails
> you've seen before to create the probability predictors.  Then you use
> those predictors to classify new email, which you really don't know
> whether they are spam or not.
>



-- 
------------------------------------------------------------------------------
Wisdom and Compassion are inseparable.
 -- Christmas Humphreys
kyle.burton@gmail.com http://asymmetrical-view.com/
------------------------------------------------------------------------------