Subscribe to DSC Newsletter

150 most popular topics discussed on AnalyticBridge since last October

Views: 243


You need to be a member of AnalyticBridge to add comments!

Join AnalyticBridge

Comment by Steffen Springer on May 11, 2010 at 6:05am
Ah, I have seen something similar during a look at reinforcement learning. Given a set of k options, the ai does not choose the best one but one with a probability correlated to the value of that option measured so far.

However: How zero is your zero probability ? When people are talking about probability they are often talking about maximum likelihood. I think we need to distinguish between 0 success (replies) and 1000 tries (clicks) and 0 success and 10 tries. We do not know how big the probability is, but given the current knowledge the former is surely less worthy than the latter.

What do I want to say:
1. I would'nt mark a topic as popular (in the sense of this thread) if it is got no replies (given a reasonable amount of clicks)
2. Featuring an article to check if it is got the power to get popular is a good (I admit: I do not know a better) strategy.
Comment by Vincent Granville on May 10, 2010 at 8:48pm
Thanks Steffen. You bring an interesting point: all articles (well.. except spam or fraud) must be given a chance, once in a while, to show up and be featured, otherwise you can end up with a very narrow set of "popular" articles that do not represent what people really like.

It's just like when our solar system was created: the Sun became the big guy in the early days, and then it continued to grow fast until it dominated everybody (Jupiter, in particular), creating a system with a very massive core (the Sun) and tiny satellites such as Jupiter etc. What I want to do, is give the little planets a chance to become a sun. From a mathematical algorithm point of view, you could say that our optimization strategy is like simulated annealing (leading to global optimum) rather than steepest ascend (leading to local optimum - that is what LinkedIn does with its group system, but I will not comply about that).

Any thoughts on this?
Comment by Steffen Springer on May 10, 2010 at 5:32am
Nice summary, Vincent. Thank you.

Some thoughts on the metric ? "Testdata wanted" for example has no replies, but has been (as far as I remember) featured once. So this topic got a lot of clicks, more than it deserved ?



On Data Science Central

© 2021   TechTarget, Inc.   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service