[WikiEN-l] Google blurb generation + Wikipedia = POVpedia

[Andrew Gray]

2009/11/16 Gregory Maxwell <gmaxwell at gmail.com>:

> For a while I thought it was just extracting text beginning at
> "Searchterm is something" looking backwards from the end of the
> article, but it seems to be more than that. Some older examples where
> I've seen this now seem to be returning different results, I don't
> know if its a timing thing or just chance.

I think it's an unfortunate collision of your search terms and
Wikipedia's preferred vocabulary. We've long standardised on "film"
not "movie", so any incidence of the latter is likely to be in direct
quotes, and is very *unlikely* to be in the lead section. Direct
quotes tend to be reviews, pro or con, so when the algorithm tries to
find extracts showing as many search terms as possible, it ends up
apparently cherry-picking these.

The effect becomes clearer when we compare the results using "film"
instead of "movie".

[the box film wikipedia]

The Box (2009 film) - Wikipedia, the free encyclopedia
"The Box is a 2009 science fiction horror film based on the 1970 short
story "Button, Button" by Richard Matheson, which was previously
adapted into an ..."

[the box movie wikipedia]

The Box (2009 film) - Wikipedia, the free encyclopedia
"And it's so rare that a movie is an F. I mean, if it's an F, it
shouldn't even be released." On the topic of the negative reaction to
The Box, Mintz blamed ..."

Note that the first has all its keywords in the header, so it shows
the first line (which contains three of them anyway). The second has
all the keywords *except* 'movie', so it looks for an extract
specifically using that word. (I don't know how it chooses that
extract, though)

We can test this by using a different keyword and seeing how it builds
the extract:

[the box mintz wikipedia]

The Box (2009 film) - Wikipedia, the free encyclopedia
"On the topic of the negative reaction to The Box, Mintz blamed the
film's ending and was quoted as saying "People really thought this was
a stinker". ..."

Might this explain the effect? No idea how to *solve* it, though...

-- 
- Andrew Gray
  andrew.gray at dunelm.org.uk