README

This PHP package for Naive Bayes works by looking at a training set and making a guess based on that set. It uses simple statistics and a bit of math to calculate the result.

What can I use this for?

You can use this for categorizing any text content into any arbitrary set of categories. For example:

is an email spam, or not spam ?
is a news article about technology, politics, or sports ?
is a piece of text expressing positive emotions, or negative emotions?

Installation

You may install Naive Bayes into your project using the Composer package manager:

composer require assisted-mindfulness/naive-bayes

Learning

Before the algorithm can do anything, it requires a training set with historical information. To teach your classifier which category the text belongs to, call the learn method:

$classifier = new Classifier();

$classifier
    ->learn('I love sunny days', 'positive')
    ->learn('I hate rain', 'negative');

Categories may be strings, integers, backed enums, or pure enums. Enum cases are normalized to their backing value or, for pure enums, their case name:

enum Sentiment: int
{
    case Negative = 0;
    case Positive = 1;
}

$classifier->learn('I love sunny days', Sentiment::Positive);

The keys returned by guess() and the value returned by most() use the normalized string or integer identifier. As with native PHP arrays, integer-like string keys such as "10" are normalized to the integer 10.

Guessing

After you have trained the classifier, you can use the prediction of which category the transmitted text belongs to, for example:

$classifier->most('is a sunny days'); // positive
$classifier->most('there will be rain'); // negative

To inspect the score for every category, use:

$classifier->guess('is a sunny days');

/*
items: array:2 [
  "negative" => 0.2461538461538462
  "positive" => 0.7538461538461538
]
*/

Probabilities are returned as Brick\Math\BigDecimal values and always add up to exactly 1 at the configured decimal scale. Any rounding remainder is assigned deterministically to the highest-ranked categories. The classifier applies Laplace smoothing to words learned in other categories and ignores words that do not occur anywhere in the training data.

Uneven

Categories have equal prior probabilities by default. If the number of learned documents should influence the result, enable uneven mode to calculate Laplace-smoothed priors from the training distribution.

$classifier
   ->uneven()
   ->guess('is a sunny days');

Precision

Returned probabilities use 16 decimal places by default. You can increase or reduce the precision through the fluent scale method:

$classifier
    ->scale(32)
    ->learn('I love sunny days', 'positive')
    ->guess('A sunny day');

The scale must be a positive integer.

Tokenizer

The algorithm utilizes a tokenizer to segment text into words. By default, it extracts Unicode letters, converts them to lowercase, and includes words longer than three characters. You can also define a custom tokenizer:

$classifier = new Classifier();

$classifier
    ->setTokenizer(function (string $string) {
        return Str::of($string)
            ->lower()
            ->matchAll('/[[:alpha:]]+/u')
            ->filter(fn (string $word) => Str::length($word) > 3);
    })
    ->learn('I love sunny days', 'positive');

The tokenizer may return any iterable of strings.

Development

Run the complete fast quality suite:

composer check

Run mutation testing with a required MSI of 90%:

composer mutation

The mutation suite measures observable behavior. Protected visibility changes are excluded, while subclass customization is covered by a dedicated behavioral test.

Wrapping up

There you have it! Even with a very small training set the algorithm can still return some decent results. For example, Naive Bayes has been proven to give decent results in sentiment analyses.

Moreover, Naive Bayes can be applied to more than just text. If you have other ways of calculating the probabilities of your metrics, you can plug those in as well.

License

The MIT License (MIT). Please see License File for more information.

assisted-mindfulness / naive-bayes

Maintainers

Package info

Statistics

Security