in reply to Re^2: Help required in RE strategy
in thread Help required in RE strategy

Yeah... but hopefully it won't degrade into fuzzy matching... it looks like matching the unmatched words with all the non-alphanum chars will get direct matches in the cases specified, and hopefully most others. That would be much cheaper and more reliable.

One could take a further step and store the word boundaries and make sure that when you make a stripped match the beginning of your would started at a boundary and the end of your word ended at a boundary, regardless of how many boundaries there were in the middle. Harder, but even more reliable (though probably unnecessary in this case).

                - Ant
                - Some of my best work - (1 2 3)

Replies are listed 'Best First'.
Re^4: Help required in RE strategy
by lima1 (Curate) on Aug 22, 2007 at 17:06 UTC
    This is not as expensive as you probably think. Assume 
    Compaq Presario Laptop Model No P440 L100 Series (PSLA0L-00U00E)

    [download]
    ... as input -> 
    00
440
100
Compaq
Laptop
Model
No
Presario
Series
PSLA

    [download]
    Then we need todo something like this: 
    ...
# get vendor id:
SELECT id FROM vendor WHERE NAME=00 or NAME=440 ...

# check all words
foreach (@word) {
  SELECT product_id FROM words WHERE vendor_id=? and word=?
  for my $id (@ids)
    $cnt{$id}++
  }
}

    [download]
    How you then analyze this match statistic depends on the data. But the editdistance is probably the wrong metric, you are right.
[reply]
[d/l]
[select]
      Don't you want that first query to be an AND... assuming you can be relatively sure of all terms matching.

                      - Ant
                      - Some of my best work - (1 2 3)

[reply]

In Section Seekers of Perl Wisdom