Hi,

I've built a prototype of this idea:

http://targetyournews.com/

Perhaps something like it could be used to give link recommendations
on del.icio.us.

This is quite preliminary work. I'm looking for people to test it out
by making submissions and targeting them as well as voting on existing
submissions.

This prototype is sort of like Reddit, but gives the submitter of a
URL control over who sees it. In other words, it is the submitter who
is responsible for the quality of recommendations for his/her
submission.

There's a catch of course. As a submitter, you need to be careful
about who will see your submission. If you spam many people with your
submission, they may just vote it down thus reducing its score.

Although the submitter has control over who sees the submission, the
score of the URL determines its ranking within the list of
recommendations for a user. So it's important that you target your
submission appropriately to maximise its score, so that it will rank
more highly as a recommendation.

We now have a real-time game. You make a submission and target it.
Targeting is based on previous submissions. You could say something
like target people who liked this earlier submission, avoid people who
disliked this submission, etc. You could have many such targeting
rules. These targeting rules combine to give users a "targeting score"
for that submission. If the targeting score is above a certain
threshold, then the user will see the submission as a recommendation.

You may change your targeting rules frequently after you have made a
submission. If your initial targeting does not appear to be working
well, you can try to tweak it. Speed matters. If you tweak it too
slowly, the score of your submission might go down quickly.

Amir

On 4/8/06, Amir Michail <[EMAIL PROTECTED]> wrote:
> Hi,
>
> Perhaps it might be interesting to allow the first person to submit a
> URL to predict who will likely be interested in that URL.
>
> For example, when submitting a URL for vim, I might indicate that
> people who bookmarked a recent but popular Windows related URL would
> likely not like this link.  But someone who bookmarked a recent but
> popular python URL and did not bookmark a recent but popular emacs URL
> would be.
>
> These predictions would be used to give URL recommendations and combat spam.
>
> Predictions that turn out to be accurate (by determining whether the
> URL was bookmarked or not for various prediction rules)  would be used
> to reward the submitter in some way.  Of course, the submitted URL
> could be recommended more aggressively based on these rules if it
> turns out that the prediction rules are turning out to be accurate
> based on initial data.
>
> A variation on this idea is to allow multiple people to make
> predictions, not just the first submitter of a url.
>
> This proposal essentially turns collaborative filtering into a prediction 
> game.
>
> Why not just use an automated algorithm?   There is a lot of
> creativity involved in coming up with prediction rules.  The search
> space is huge.
>
> Moreover, it might be fun to see how well your prediction rules work out.
>
> Amir
>
_______________________________________________
discuss mailing list
[email protected]
http://lists.del.icio.us/cgi-bin/mailman/listinfo/discuss

Reply via email to