text-similarity-users

[text-similarity-users] some clarifications about Text::similarity

[Yashar M. <yas...@ya...>]

Dear Ted

I'm using your package in some of my experiments and in order to cite it I need few clarifications :

while "text_similarity.pl --type=Text::Similarity::Overlaps file1 file2" is executed a normalized measure is obtained. what I understood from your documentation this measure is raw normalized (F-measure = 2 * precision * recall / (precision + recall)), is that right?
while "text_similarity.pl --type=Text::Similarity::Overlaps --no-normalize file1 file2" is executed, the output would be a simple raw score of overlap not lesk raw score? is it right?is there any way in which by using text_similarity.pl an reach the lesk measure through defining any option? (im aware that by defining verbose option we could get all measures but is there any way that directly lead us to lesk measure).Thanks in advacne for your reply and help.

Best regards
Yashar.





      

Re: [text-similarity-users] some clarifications about Text::similarity

[Ted P. <dul...@gm...>]

Hi Yashar,

Thanks for your questions - see my responses inline...

On Tue, Jan 20, 2009 at 9:31 AM, Yashar Mehdad <yas...@ya...> wrote:
> Dear Ted
>
> I'm using your package in some of my experiments and in order to cite it I
> need few clarifications :
>
> while "text_similarity.pl --type=Text::Similarity::Overlaps file1 file2" is
> executed a normalized measure is obtained. what I understood from your
> documentation this measure is raw normalized (F-measure = 2 * precision *
> recall / (precision + recall)), is that right?

Correct. Consider the following example...

ted@ted-desktop:~$ text_similarity.pl test1 test2
--type=Text::Similarity::Overlaps --no-normalize
5

ted@ted-desktop:~$ text_similarity.pl test1 test2
--type=Text::Similarity::Overlaps
0.555555555555556

ted@ted-desktop:~$ more test1
this is test1 i am happy he is not

ted@ted-desktop:~$ more test2
this is test2 i am hungry she is sad

no-normalize for the second result shows that 5 words have matched
(without regard to order
or length of phrase)

>
> while "text_similarity.pl --type=Text::Similarity::Overlaps --no-normalize
> file1 file2" is executed, the output would be a simple raw score of overlap
> not lesk raw score? is it right?

Correct! There is no "bonus" for phrasal matching in the overlap scoring.

>
> is there any way in which by using text_similarity.pl an reach the lesk
> measure through defining any option? (im aware that by defining verbose
> option we could get all measures but is there any way that directly lead us
> to lesk measure).

Not from the command line, however, you could edit Overlaps.pm to just
output lesk....

Here's the relevant snippet, where I've added comments...

if ($self->verbose) {
#	    print " Raw score: $score\n";
#	    print " Precision: $prec\n";
#	    print " Recall   : $recall\n";
#	    print " F-measure: $f\n";
#	    my $dice = 2 * $score / ($wc1 + $wc2) ;
#	    print " Dice     : $dice\n";

#	    my $e = 1 - $f;
#	    print " E-measure: $e\n";

#	    my $cos = $score / sqrt ($wc1 * $wc2);
#	    print " Cosine   : $cos\n";
	

            my $lesk = $raw_lesk/ ($wc1 * $wc2);
#	    print " Raw lesk : $raw_lesk\n";
	    print " Lesk     : $lesk\n";


I know that's a bit messy, but should be a fairly easy fix in the
short term at least...

I hope this helps!
Ted

>
> Thanks in advacne for your reply and help.
>
> Best regards
> Yashar.
>
>
>
>
> ------------------------------------------------------------------------------
> This SF.net email is sponsored by:
> SourcForge Community
> SourceForge wants to tell your story.
> http://p.sf.net/sfu/sf-spreadtheword
> _______________________________________________
> text-similarity-users mailing list
> tex...@li...
> https://lists.sourceforge.net/lists/listinfo/text-similarity-users
>
>



-- 
Ted Pedersen
http://www.d.umn.edu/~tpederse