[Tagdb] Tag Relevance

[anand]

On 2/8/07, Nitin Borwankar <nitin at borwankar.com> wrote:
>
> anand wrote:
>
> > Hi,
> >
> > I have been involved in designing an information retrieval system that
> > takes tag meta-data into account as well. I figured out two approaches
> > to rank items based on the tag information:
> >
> >    1. Purely based on the number of users that have tagged that item.
> >       Say item1 has been tagged by 10 users as tag1 and item2 by 20
> >       users as tag1, then item1 would end-up having a higher ranking.
> >
>
> Did you actually mean item2 would have a higher ranking since it had
> more taggers.


Yes.

>   1.
> >
> >
> >    2. Based on the relative ranking of a tag when the collective
> >       wisdom of the crowd is taken into account. To further the
> >       explanation consider 2 items. Item1 has tags tag1(by 100 users)
> >       and tag2(by 10 users). Item2 has tag2(by 10 users). When a user
> >       searches for tag2, by approach 1 (above) ietm1 and item2 would
> >       be comparable as both have 10 users tagging it as tag2. But in
> >       this approach, item2 would be more relevant as for item1
> >       although 10 users have tagged it as tag2, 100 users have (taking
> >       collective wisdom into account) tagged it as tag1. Hence tag2
> >       for item1 is relatively less relevant than tag2 for item2.
> >
>
>
> Interesting!
>
> So here you are calculating something like the following normalized
> relevance ?


Yeah..something similar to that. Which makes more sense to you?

number of uses of  requested tag
> ---------------------------------------------------------
> Sum over all tags ( number of instances of tag )
>
>
> Nitin
>
>
> >   1.
> >
> >
> >
> > In my opinion, approach 2 seems more reasonable as it takes into
> > account how people are collectively describing an object. But some
> > people beg to differ as they reason that tags need to be treated
> > individually and are independent of the other tags.
> >
> > Which do you think would yield better results? Your thoughts on the
> > same? Any other approach that you might have used?
> >
> > Thanks,
> > Anand.
> >
> >------------------------------------------------------------------------
> >
> >_______________________________________________
> >Tagdb mailing list
> >Tagdb at lists.tagschema.com
> >http://lists.tagschema.com/mailman/listinfo/tagdb
> >
> >
>
>
> --
> Nitin Borwankar
> Find, Learn, Act ....
> Greener, the search engine for the planet
> http://greener.com
> nitin at borwankar.com
> 510-872-7066
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.tagschema.com/pipermail/tagdb/attachments/20070208/48e3645d/attachment.htm