If each cleanup action is different, consider putting that code in a subroutine to simplify the while loop. For example

#!perl
use strict;

my %mycorpus = (
  a => "date:#20180101# title:#cat dog# text:#sheep sheep sheep sheep#
+" ,
  b => "date:#20180101# comment:#woof woof#",
  c => "date:#20180101# title:#cow puppy# text:#pig pig pig#",
);

my %counts;
foreach my $filename (sort keys %mycorpus) {
  my $date;

  if ($mycorpus{$filename} =~ /date:#(\d{8})#/g){ 
    $date = $1;  
  } else {
    die "No date found filename : $filename => $mycorpus{$filename}";
  }
  
  while ($mycorpus{$filename} =~ /(title|text|comment):#(.*?)#/g){
    my $dataset = cleanup($1,$2);
    my @words = split /\W+/, $dataset;

    foreach my $word (@words){

      if ($word =~ /(\w+)/gi){
        $word =~ tr/A-Z/a-z/;
        $counts{$date}{$word}++;
       #$word_types{$word}++;
       #$overallcounts{$date}++;
      }
    }   
  }  
}

sub cleanup {
  my ($key,$value) = @_;
  if ($key eq 'title'){
     #Actions usually performed here which clean the titles 
     #(i.e. substituting certain characters)
  }
  if ($key eq 'text'){
     #Actions usually performed here which clean the text 
  }
  if ($key eq 'comment'){
     #Actions usually performed here which clean the comments 
  }
  return $value;
}

use Data::Dumper;
print Dumper \%counts;
[download]