> The point of my post was that people *do not* have to use the advanced
> search screen to do this "uberdweeb" kind of searching. What they can do is
> click on the link that they see in the bibliographic record after they do a
> keyword search. Therefore, they can search "assassinations italy" and find a
> record with the subjects:
> * Assassination -- History -- Italy-- 20th century.
> * Assassination -- History -- Italy-- 21st century.
> * Murder -- History -- Italy-- 20th century.
> * Murder -- History -- Italy-- 21st century.
> * Political violence -- History -- Italy-- 20th century.
I'm totally with you on this. It's just that the numbers seemed really skewed.
One of the weaknesses of most implementation of click through subject
headings is that there are too many unique headings. For example, all
of the headings above appear to be unique in your catalog which limits
their utility as they all point to only one record.
Last year, I toyed around with using subject headings as access
points. My experiments indicated that a good way to deal with them was
to get rid of the precoordination (because concepts are often spread
across multiple headings), remove duplicate terms, and then perform
controlled vocab keyword relevancy against that.
For example, the example you have above would be reduced to
["Assassination", "Murder", "Political violence", "History", "Italy",
"20th century", "21st century"]. Since 650|a is more important than
the other subfields, it makes sense to give more weight to matches
against |a. Deduplication is important because |v, |x, |y, and |z are
repeated more often than |a -- this has the effect of boosting
unimportant terms and and deemphasizing the real access points.
Unfortunately, I had to abandon my work for reasons I can't go into
here. However, I believe the technique is sound and think that a good
way to implement subject searching in a catalog would to perform an
AJAX call upon mouseover of subject headings to show the 10 most
relevant retrievals. Clicking would bring the entire list.
It is a little dangerous to infer effectiveness of searches by their
relative popularity. Someone sent me an off list message suggesting
that subject searches are 80 times more popular than title searches in
your catalog because people using the latter find what they want right
away and quit, while those choosing the former strategy just spin in
circles. There is a kernel of truth in every joke....
kyle
Received on Fri Feb 15 2008 - 11:58:18 EST