Instead of nested-looping over the words looking for pairs, just count up the number of key words that are found, and go from there. The (working/tested) code below first parses the data into a data structure using XML::Simple (uncomment the Dumper to see what it looks like). Then it "prompts" (i have it hardcoded as a constant for testing) for the input, which it then hashes up into %searchwords. Now, you can start going through each MS_X item, and take the <words> section and split it into words, then count how many of those words exist in the %searchwords dictionary. Just add if($score>=2){...} logic inside the loop.

use strict;
use warnings;

my $s = do {local $/=undef; <DATA>};
use XML::Simple;
my $data = XMLin("<opt>$s</opt>", KeepRoot => 0);
#use Data::Dumper;
#print Dumper $data;

my $search_string = "dog cat cow";  # from input
my %searchwords = map { $_ => undef } split ' ', $search_string;
while( my ($k, $h) = each %$data ){
  my $score = scalar grep( exists $searchwords{$_}, split(/, /, $h->{w
+ords}) );
  # do whatever based on score.
}

__DATA__
<MS_1>
  <loc>c:\data\cat.xml</loc>
  <words>dog, cat, fish, bird</words>
</MS_1>

<MS_2>
  <loc>c:\data\cow.xml</loc>
  <words>dog, cat, fish, bird, cow, goat</words>
</MS_2>

<MS_3>
  <loc>c:\data\snake.xml</loc>
   <words>dog, cat, fish, bird, snake, orange</words>
</MS_3>
[download]

One comment on your attempt -- it looks like you were trying to re-read <DATA> inside in the inner loop .. you would have to seek to the beginning tfor that to work, but should just store it in a variable before the loops -- other wise it's a waste of expensive I/O to keep opening the file up.