use strict;
use warnings;

use Tree::Builder;
use Data::Dumper;

my @items = qw(
    Devices_CLI_Inspection
    Devices_CLI_Inspection_L2
    Devices_CLI_Inspection_L2_FilterDatabase_General
    Devices_CLI_Inspection_L2_SpanningTree
    Devices_CLI_Inspection_L2_SpanningTree_MST
    Devices_CLI_Inspection_L2_VLAN_General
    Devices_CLI_Inspection_System_Config
);

my $top = shift @items;

my @items2 = @items;
for (@items2) { 
    s/^Devices_CLI_Inspection//;
    s{_}{/}g;
}

my $tb = Tree::Builder->new();
$tb->add($_) for @items2;

my %hash=$tb->getTree();
print Dumper(\%hash);

__END__

$VAR1 = {
          '' => {
                  'System' => {
                                'Config' => {}
                              },
                  'L2' => {
                            'VLAN' => {
                                        'General' => {}
                                      },
                            'SpanningTree' => {
                                                'MST' => {}
                                              },
                            'FilterDatabase' => {
                                                  'General' => {}
                                                }
                          }
                }
        };
[download]

Have a look at Text::Trie.

use Data::Dump qw(dump);
use Text::Trie qw(Trie walkTrie);
my @trie = Trie qw(
Devices_CLI_Inspection
Devices_CLI_Inspection_L2
Devices_CLI_Inspection_L2_FilterDatabase_General
Devices_CLI_Inspection_L2_SpanningTree
Devices_CLI_Inspection_L2_SpanningTree_MST
Devices_CLI_Inspection_L2_VLAN_General
Devices_CLI_Inspection_System_Config
);

dump(\@trie);

walkTrie sub {print("[",shift,"]")}, sub {print(shift)},
  sub {print "->"}, sub {print ","}, sub {print "("}, 
  sub {print ")"}, @trie;
[download]

You could try splitting on '_' and creating hashes of hashes of hashes of ...

Why does VLAN_General get its own bucket. It seems as if it should go with the bucket right above it.

ya unfortunately the order above is completely random...but I think your idea of multiple hashes might do the trick. I'll give it a try.

Thanks!

Hashes will sort out uniqueness independently of how random the input is.

The only problem you'll run into is that things like System_Config would be split into two (unlike what you seem to want according to your sample), even if you stop splitting after 4 underscores (which is the minimum depth required for getting a separate L2 entry). So you'd have to invent some additional logic to determine how deep to split...