mirror of
https://github.com/pixelfed/pixelfed.git
synced 2024-11-30 10:13:16 +00:00
178 lines
3.3 KiB
PHP
178 lines
3.3 KiB
PHP
<?php
|
|
|
|
namespace App\Util\Lexer;
|
|
|
|
use Brick\Math\BigDecimal;
|
|
use Illuminate\Support\Collection;
|
|
use Illuminate\Support\Str;
|
|
|
|
class Classifier
|
|
{
|
|
/**
|
|
* @var ?callable(string): array<int, string>
|
|
*/
|
|
private $tokenizer;
|
|
|
|
/**
|
|
* @var array<string, array<string, int>>
|
|
*/
|
|
private array $words = [];
|
|
|
|
/**
|
|
* @var array<string, int>
|
|
*/
|
|
private array $documents = [];
|
|
|
|
private bool $uneven = false;
|
|
|
|
/**
|
|
* @param callable(string): array<int, string> $tokenizer
|
|
*/
|
|
public function setTokenizer(callable $tokenizer): void
|
|
{
|
|
$this->tokenizer = $tokenizer;
|
|
}
|
|
|
|
/**
|
|
* @return Collection<int, string>
|
|
*/
|
|
public function tokenize(string $string): Collection
|
|
{
|
|
if ($this->tokenizer) {
|
|
/** @var array<int, string> */
|
|
$tokens = call_user_func($this->tokenizer, $string);
|
|
|
|
return collect($tokens);
|
|
}
|
|
|
|
return Str::of($string)
|
|
->lower()
|
|
->matchAll('/[[:alpha:]]+/u');
|
|
}
|
|
|
|
/**
|
|
* @return $this
|
|
*/
|
|
public function learn(string $statement, string $type): self
|
|
{
|
|
foreach ($this->tokenize($statement) as $word) {
|
|
$this->incrementWord($type, $word);
|
|
}
|
|
|
|
$this->incrementType($type);
|
|
|
|
return $this;
|
|
}
|
|
|
|
/**
|
|
* @return Collection<string, string>
|
|
*/
|
|
public function guess(string $statement): Collection
|
|
{
|
|
$words = $this->tokenize($statement);
|
|
|
|
return collect($this->documents)
|
|
->map(function ($count, string $type) use ($words) {
|
|
$likelihood = $this->pTotal($type);
|
|
|
|
foreach ($words as $word) {
|
|
$likelihood *= $this->p($word, $type);
|
|
}
|
|
|
|
return (string) BigDecimal::of($likelihood);
|
|
})
|
|
->sortDesc();
|
|
}
|
|
|
|
public function most(string $statement): string
|
|
{
|
|
/** @var string */
|
|
return $this->guess($statement)->keys()->first();
|
|
}
|
|
|
|
/**
|
|
* @return self
|
|
*/
|
|
public function uneven(bool $enabled = false): self
|
|
{
|
|
$this->uneven = $enabled;
|
|
|
|
return $this;
|
|
}
|
|
|
|
/**
|
|
* Increment the document count for the type
|
|
*/
|
|
private function incrementType(string $type): void
|
|
{
|
|
if (! isset($this->documents[$type])) {
|
|
$this->documents[$type] = 0;
|
|
}
|
|
|
|
$this->documents[$type]++;
|
|
}
|
|
|
|
/**
|
|
* Increment the word count for the given type
|
|
*/
|
|
private function incrementWord(string $type, string $word): void
|
|
{
|
|
$ignored = config('autospam.ignored_tokens');
|
|
if(!$ignored) {
|
|
$ignored = ['the', 'a', 'of', 'and'];
|
|
} else {
|
|
$ignored = explode(',', $ignored);
|
|
}
|
|
if ($type == 'spam' && in_array($word, $ignored)) {
|
|
return;
|
|
}
|
|
if (! isset($this->words[$type][$word])) {
|
|
$this->words[$type][$word] = 0;
|
|
}
|
|
|
|
$this->words[$type][$word]++;
|
|
}
|
|
|
|
/**
|
|
* @return float|int
|
|
*/
|
|
private function p(string $word, string $type)
|
|
{
|
|
$count = $this->words[$type][$word] ?? 0;
|
|
|
|
return ($count + 1) / (array_sum($this->words[$type]) + 1);
|
|
}
|
|
|
|
/**
|
|
* @return float|int
|
|
*/
|
|
private function pTotal(string $type)
|
|
{
|
|
return $this->uneven
|
|
? ($this->documents[$type] + 1) / (array_sum($this->documents) + 1)
|
|
: 1;
|
|
}
|
|
|
|
public function export()
|
|
{
|
|
$words = $this->words;
|
|
$words = collect($words)
|
|
->map(function($w) {
|
|
arsort($w);
|
|
return $w;
|
|
})
|
|
->all();
|
|
return json_encode([
|
|
'_ns' => 'https://pixelfed.org/ns/nlp',
|
|
'_v' => '1.0',
|
|
'documents' => $this->documents,
|
|
'words' => $words
|
|
], JSON_PRETTY_PRINT|JSON_UNESCAPED_SLASHES);
|
|
}
|
|
|
|
public function import($documents, $words)
|
|
{
|
|
$this->documents = $documents;
|
|
$this->words = $words;
|
|
}
|
|
}
|