addressalign-toparrow-leftarrow-rightbackbellblockcalendarcameraccwchatcheckchevron-downchevron-leftchevron-rightchevron-small-downchevron-small-leftchevron-small-rightchevron-small-upchevron-upcircle-with-checkcircle-with-crosscircle-with-pluscrosseditemptyheartexportfacebookfolderfullheartglobegmailgoogleimageimagesinstagramlinklocation-pinmagnifying-glassmailminusmoremuplabelShape 3 + Rectangle 1outlookpersonplusprice-ribbonImported LayersImported LayersImported Layersshieldstartickettrashtriangle-downtriangle-uptwitteruseryahoo

RE: [newtech-1] Web Commenting Systems

From: Alex O.
Sent on: Monday, March 3, 2008 11:30 PM
Hello Asif -

I think you really need to flush out what your goals are with this project.  This is generally a VERY hard problem and it will be easier to give you advice if you can give us a little more information.

From reading your message I think (vaguely) you are trying to create a ranking on a set of descriptions (comments) of a single item (car).  This is a very interesting problem that is part of a larger group of problems of using the power of the masses that I would love to discuss but let me stick to something practical for now.

A starting point to removing the bias from vote inflation can be bayes averaging.  Bayes averaging to calculate rank will be sufficient for simple rating systems and you will be able to extend the formulas to fit your own goals.  I think time cycles will be involved in car ratings simply as a result of the car release cycles so you will most likely have to model for that bias as well.

This is a statistical science for the most part and you should be able to come up with different models for different thresholds and use different models accordingly.  Bayes averaging and different factors based on what you are trying to achieve would work well in different scenarios.  In a case such as a car is brand new, a ton of comments have flooded in the first day BUT people have not voted whether or not the comments are useful you should can use a different metric to rank comments such as previous community participation (e.g. give posters with more useful comments higher ranking).

My advice though if you are not trying to solve all these problems but looking for a practical method is to look at what some of the big players in this area are doing.  Amazon's success can be at least partially attributed to their comment system, so they are a great innovator to follow - and you can be sure that with Amazon's scale they have tested various techniques in a statistically meaningful way.  For their scenario the comment system with did you find this useful works best.  Microsoft recently switched from using a 5 star rating system for music to a like/dislike system, according to them it is a better statistical indicator of musical likes.

Overall though this is an unsolved problem and we can only come up heuristic methods (at least for now).  But if you can come up with a great way to rank your comments I am sure you will find quick celebrity (what is Google after-all if not a scalable ranking algorithm).

Sorry if I am all over the place but I am really not sure what you are trying to accomplish so not sure which direction to point you in.  Post some more information on what you're really trying to achieve and I am sure you'll find comments more helpful.

Alex

-----Original Message-----
From: [address removed] [mailto:[address removed]] On Behalf Of Asif Youssuff
Sent: Monday, March 03,[masked]:25 PM
To: [address removed]
Subject: [newtech-1] Web Commenting Systems

Hey all,

I'm working on a freelance project on the side, and I've been tasked
with creating a commenting system for it.

Let me give you guys an idea of what the comments are for, and what I'd
like to accomplish...

The site is basically going to be a database of items, for simplicity's
sake, let's call them cars.

The site will allow users to tag these items for their own relevancy (as
well as others).

However, commenting will be needed to get a (rough) aggregate idea of
what the users think of the car.

What problems am I trying to solve?

Vote inflation: We don't want any one user to be able to influence the
tally to an immense degree (by posting[masked] times), but we want users
to comment as often as they'd like.

Relevancy: The user can go through and view all comments, however, a
casual user will likely only look at the top view comments.

How can we make it so that the "best comments" are listed as being
representative of the car? Listing by date won't work, and listing by
amount of comments submitted doesn't work either.

Timeliness: The comment should not be outdated, this feeds into the
relevancy problem.

I'm sure that there are issues that I'm not covering here, but those are
what I came up with in a bit of armchair modeling.

I also thought about a cut-down slashcode moderation type system, but
that may be too complex, and the amount of users needed to sustain a
good moderation system might not be present with this site.

What am I looking for?

Code, preferably open source, so I can recreate, or reuse it.

Philosophies on how to do it right.

Any other ideas. I want to do it well, and it's not an easy problem to
solve /well/.

-Asif



--
Please Note: If you hit "REPLY", your message will be sent to everyone on this mailing list ([address removed])
http://newtech.me...­
This message was sent by Asif Youssuff ([address removed]) from NY Tech Meetup.
To learn more about Asif Youssuff, visit his/her member profile: http://newtech.me...­
To unsubscribe or to update your mailing list settings, click here: http://www.meetup...­

Meetup.com Customer Service: [address removed]
632 Broadway New York NY 10012 USA

Our Sponsors

People in this
Meetup are also in:

Sign up

Meetup members, Log in

By clicking "Sign up" or "Sign up using Facebook", you confirm that you accept our Terms of Service & Privacy Policy