Please suggest.

What have you tried? Did you consider using a regular expression with m?

my $string = "stash_ST1.2.3.4_1.ROLLBACK.check.20170228-101051.435944.
+txt";
# find one or more not "." between pair of "."s following word ROLLBAC
+K
my ($word) = $string =~ m/ROLLBACK\.([^\.]+)\./; 
print "$word\n";  # prints "check"
[download]

UPDATE: Note that the code that this would require is the same as scorpio17's except for the regex. He implements a different definition of 'word'.

Does the start of the string before ROLLBACK vary? If so, what other kinds of prefixes are there? Or is "ROLLBACK" always in the string before the part you're interested in? (If it is, you don't even need a regex; a simple index call followed by substr will work.)

ROLLBACK is always a part of string and I need only the word which follows ROLLBACK.

If that's all it is, you can try this instead of a regular expression:

#!/usr/bin/perl
use v5.10;
my $string = "stash_ST1.2.3.4_1.ROLLBACK.check.20170228-101051.435944.
+txt";
# Find "ROLLBACK" from the end of string, in case there's more than on
+e of
# them.
my $i = rindex($string, 'ROLLBACK.');
if ($i >= 0) {
    $i += length('ROLLBACK.'); 
    # Now $i points to the character after the full stop.
    # Find the position of the next full stop after it.
    my $j = index($string, '.', $i);
    if ($j >= 0) {
        my $word = substr($string, $i, $j - $i);
        say $word;
    } else {
        # No full stop after ROLLBACK -- signal an error?
    }
} else {
    # ROLLBACK not found! Signal an error?
}
[download]

However, scorpio17's regular expression is better, because you can see the input string's expected structure more clearly and the regex solution is also shorter!

Not that fancy, but straightforward.

#!/usr/bin/perl -w
use strict;
my $string = "stash_ST1.2.3.4_1.ROLLBACK.check.20170228-101051.435944.
+txt";
my $action = '';
my @parts = split(/\./, $string); # split on dot
for (0..$#parts){
    # slot after ROLLBACK
    $action = $parts[$_+1] if $parts[$_] eq "ROLLBACK";
}    
print "$action\n";
[download]

I'm sure this can be golfed down a bit.

FWIW, split(/\./, $string) can be rewritten as split('.', $string) which avoids using a regular expression.

It also avoids the result, though.

my $s = 'a.b.c';
say "1: $_" for split /\./, $s;
say "2: $_" for split '.', $s;
[download]

($q=q:Sq=~/;[c](.)(.)/;chr(-||-|5+lengthSq)`"S|oS2"`map{chr |+ord
}map{substrSq`S_+|`|}3E|-|`7**2-3:)=~y+S|`+$1,++print+eval$q,q,a,
[download]