my $positive = {
    word1 => 2,
    word2 => 4,
    word3 => 1,
};
my $negative = {
    word4 => 3,
    word5 => 1,
};
[download]

my $categorizer = Algorithm::NaiveBayes->new;

$categorizer->add_instance(
    attributes => $positive,
    label => 'positive');

$categorizer->add_instance(
    attributes => $negative,
    label => 'negative');

$categorizer->train;
[download]

my $sentence1 = {
    wordA => 2,
    wordB => 1,
};

my $probability = $categorizer->predict(attributes => $sentence1);

if ($probability->{'positive'} > 0.5) {
    # sentence1 probably positive
}
elsif ($probability->{'negative'} > 0.5) { 
    # sentence1 probably negative
}
[download]

Thanks a lot for your help. I will read the book you recommend! Best wished!!

hi~ I have already read the chapter "Categorization and Extraction" in Advanced Perl Programming and almost get the idea how to do my work. Here are some problems: 1.in the book the author has invert the document into the hash of words and weights. using the code below:

sub invert_string {
       my ($string, $weight, $hash) = @_;
       $hash->{$_} += $weight for
            grep { !$StopWords{$_} }
            @{words(lc($string))};
    }
[download]

But,I have already do the stem and stop word in advance so I think the code you wrote:

my $sentence1 = {
    wordA => 2,
    wordB => 1,
};
[download]

has the same function, is there any difference?

2. If I have hundreds of sentence to make prediction, how can I invert all the sentence into hash variable?

3.I can not quite get the function of the code:

sub invert_item {
        my $item = shift;
        my %hash;
        invert_string($item->{title}, 2, \%hash);
        invert_string($item->{description}, 1, \%hash);
        return \%hash;
    }
[download]

is that true because I do not need to separate the weight of title and contend, so I can ignore this step?

4.here is the code in the book to train analyzer:

 #!/usr/bin/perl

    use XML::RSS;
    use Algorithm::NaiveBayes;
    use Lingua::EN::Splitter qw(words);
    use Lingua::EN::StopWords qw(%StopWords);

    my $nb = Algorithm::NaiveBayes->new(  );

    for my $category (qw(interesting boring)) {
        my $rss = new XML::RSS;
        $rss->parsefile("$category.rdf");
        $nb->add_instance(attributes => invert_item($_),
                          label      => $category) for @{$rss->{'items
+'}};
    }

    $nb->train; # Work out all the probabilities
[download]

I don't understand the function of:

 my $rss = new XML::RSS;
        $rss->parsefile("$category.rdf");
       $nb->add_instance(attributes => invert_item($_),
                          label      => $category) for @{$rss->{'items
+'}};
    }
[download]

If I have ignore the invert_item step, what should I write to take place of "@{$rss->{'items'}}"

5. all the codes you wrote above should be written in one perl document or their have to be written separately and quote each other by name?

Your reply will be surely helpful. Thank you so much!!

hi~ I have already read the chapter "Categorization and Extraction" in Advanced Perl Programming and almost get the idea how to do my work. But I still have some questions, I really wish you can help me to solve them, Here are the questions:

1.when I train categories, the code you wrote is

my $positive = {
    word1 => 2,
    word2 => 4,
    word3 => 1,
};
[download]

is the code "word1 => 2" mean the number of times word1 appear in positive sentence? If I have 100 sentences taken as training sentence, then I need to produce hash for all the words in these sentence? Is there any easier way for me to train all the sentence?

2.in the book the author has invert the document into the hash of words and weights. using the code below:

sub invert_string {
       my ($string, $weight, $hash) = @_;
       $hash->{$_} += $weight for
            grep { !$StopWords{$_} }
            @{words(lc($string))};
    }
[download]

But,I have already do the stem and stop word in advance so I think the code you wrote:

my $sentence1 = {
    wordA => 2,
    wordB => 1,
};
[download]

has the same function, is there any difference? If I have 100 training sentence, do I need to type all the different words in these training sentence using the code above?

3. If I have hundreds of sentence to make prediction, how can I invert all the sentence into hash variable?

4.I can not quite get the function of the code:

sub invert_item {
        my $item = shift;
        my %hash;
        invert_string($item->{title}, 2, \%hash);
        invert_string($item->{description}, 1, \%hash);
        return \%hash;
    }
[download]

is that true because I do not need to separate the weight of title and contend, so I can ignore this step?

5.here is the code in the book to train analyzer:

 #!/usr/bin/perl

    use XML::RSS;
    use Algorithm::NaiveBayes;
    use Lingua::EN::Splitter qw(words);
    use Lingua::EN::StopWords qw(%StopWords);

    my $nb = Algorithm::NaiveBayes->new(  );

    for my $category (qw(interesting boring)) {
        my $rss = new XML::RSS;
        $rss->parsefile("$category.rdf");
        $nb->add_instance(attributes => invert_item($_),
                          label      => $category) for @{$rss->{'items
+'}};
    }

    $nb->train; # Work out all the probabilities
[download]

I don't understand the function of:

 my $rss = new XML::RSS;
        $rss->parsefile("$category.rdf");
       $nb->add_instance(attributes => invert_item($_),
                          label      => $category) for @{$rss->{'items
+'}};
    }
[download]

If I have ignore the invert_item step, what should I write to take place of "@{$rss->{'items'}}"

6. all the codes you wrote above should be written in one perl document or their have to be written separately and quote each other by name?

Your reply will be surely helpful. Thank you so much!!

Say you have three files: positive, negative, and the sentences to test. They are already prepared and are in this format:

wordA wordB wordC
wordD wordA wordE wordF
[download]

To train you would feed the first two files in:

my $pos_file = '/path/to/positive.txt';
my $neg_file = '/path/to/negative.txt';

my $categorizer = Algorithm::NaiveBayes->new;
my $fh;

open($fh,"<",$pos_file) or die "Could not open $pos_file: $!";
while (my $sentence = <$fh>) {
    chomp $sentence;
    my @words = split(' ',$sentence);
    my %positive;
    $positive{$_}++ for @words;
    $categorizer->add_instance(
        attributes => \%positive,
        label => 'positive');
}
close($fh);

open($fh,"<",$neg_file) or die "Could not open $neg_file: $!";
while (my $sentence = <$fh>) {
    chomp $sentence;
    my @words = split(' ',$sentence);
    my %negative;
    $negative{$_}++ for @words;
    $categorizer->add_instance(
        attributes => \%negative,
        label => 'negative');
}
close($fh);

$categorizer->train;
[download]

You can then feed the third file in:

my $sentence_file = '/path/to/sentence.txt';

open($fh,"<",$sentence_file) or die "Could not open $sentence_file: $!
+";
while (my $sentence = <$fh>) {
    chomp $sentence;
    my @words = split(' ',$sentence);
    my %test;
    $test{$_}++ for @words;
    my $probability = $categorizer->predict(attributes => \%test);
    # ...
    # do what you need with $probability
}
close($fh);
[download]

The Synopsis section of the Algorithm::NaiveBayes docs seems fairly straight-forward. While I haven't tested the code below, I believe it should work as intended.

Hope that helps.

Hi! Thank you for your code again. Can you please tell me the function of Data::Dumper? if I want to classify each training sentence in to certain category, such as: revenue, cost, profit and so on, since the same word will have different tone in different environment, for example: the word increase would be positive, if it appears in the sentence about revenue, but it would be negative, if it appears in the sentence about cost. How can I modify the code to implement the function? Your reply will be helpful to me. Thank you!

I have run the code, the result is like this:

Prediction: HASH(0x7f8f540c4c00) - /Users/Agnes/Documents/process_sentence.txt

Prediction: HASH(0x7f8f540c2380) - /Users/Agnes/Documents/test.txt

my code is as follow:

#!/usr/bin/perl

use Algorithm::NaiveBayes;
use Data::Dumper;

$| = 1;
$Data::Dumper::Deepcopy = 1;
$Data::Dumper::Sortkeys = 1;

my %training_files = 
    (   positive => q{/Users/Agnes/Documents/positive.txt},
        negative => q{/Users/Agnes/Documents/negative.txt},
        neutral  => q{/Users/Agnes/Documents/neutral.txt},  );

my @test_files = 
    (   q{/Users/Agnes/Documents/process_sentence.txt},
        q{/Users/Agnes/Documents/test.txt},  );

my $nb = Algorithm::NaiveBayes->new( purge => 0, );

foreach my $k ( keys %training_files ) {
    local $/;
    open my $inf, q{<}, $training_files{$k} or die $!;
        my $line = <$inf>;
    close $inf;
    $nb->add_instance( 
        attributes => str_to_array( $line ), 
        label => [ $word ], 
    );
}

$nb->train;

foreach my $tf ( @test_files ) {
    local $/;
    open my $inf, q{<}, $tf or die $!;
        my $line = <$inf>;
    close $inf;
    my $result = $nb->predict(
                        attributes => str_to_array( $line ),
                    );
    print qq{Prediction: $result - $tf\n};
}

sub str_to_array {
    my ($str) = @_;
    my %attr;
    foreach my $word ( split /\s|[\(\)!?.,:;]/, $str ) {
        $attr{$word}++;
    }
    return \%attr;
}
[download]

I s there any problem in it?

Thank you so much! I will try this code and report my result to you!! Have a good night!!