I started writing this as a personal mail to Tim
Baker, then as it became bigger and bigger I realised
how it would probably be of interest to the whole
community, so I send it to the discussion list and in
copy to Tim.
***
I just came back from the vacations and I find the two
new jewels of delicious:
the search engine and the reccomendation system.
Really cool.
I have some questions though.
What is the alghorithm behind them.
In particular I tested the search engine, and found
that if I looked for two quite different tags
'religion' and 'folksonomy' I would get quite some
results, but none that contained both the keywords.
Not just not in the search page (which I gather gets
the most heavy -high weight- tags), but also in the
individual page. Infact I would often get one of the
result, look at all the person that has bookmarked
that entry, look for one of the two keywords, and
either one or the other would be absent.
Is this how is it suppose to behave or is it a bug?
The reason why I chose religion and folksonomy is
because I know there is a page that has been tagged
with both those tags, but such that no one person used
both the tags, and I wanted to make sure the search
engine was searching on the whole tag cloud (as
implied by Brian in the discussion group entry of some
time ago).
It might even be that you are using the search on the
hipercube and it is nearer a result that has a high
value for folksonomy and nil for religion, than one
that has a non nil value for both but not a high value
on either. If this is the case we should think about
it, maybe adding the symbol '+' before a keyword to
clarify that the keyword MUST be present might be a
good idea.
Now speaking about the reccomendation system.
It is REALLY GOOD. I have already found some very good
pages. In fact I just went to an article I was very
interested to, looked for related and got a bunch of
articles, EACH important and EACH new. WOW.
Now some questions.
1. Again what algorithm does it uses? Joshua sais, is
the same algorithm as before but I could not remember
any post describing it in the last months.
2. Could the results be offered in the same way as the
search results are given:
Name of URL, number of people using it, most relevan
tags.
In this way it would be very easy to navigate
delicious going from one link to the relevant, to the
next relevant and so on. Some group of URL will make
small islands, and some will make bigger islands, but
users will have the possibility to explore the whole
island.
3. if the reccomandation system uses a distance
between urls, are we sure that getting the first n is
the most wise way to proceede? Seem to me that it
should give ALL the results below a certain
threashold. And if the number of results is less than
x, than (and only then) the best n anyway.
Said all this I really like the reccomandation system,
and it seem to me that it has the potentiality to
bring delicious to a whole other level.
Regards,
Pietro Speroni
http://www.pietrosperoni.it
Start your day with Yahoo! - make it your home page
http://www.yahoo.com/r/hs
___
discuss mailing list
discuss@del.icio.us
http://lists.del.icio.us/cgi-bin/mailman/listinfo/discuss
