如何使用php-ai/php-ml与K-means聚类将句子分组为相似集合？

我尝试将句子数组按句中单词的相似度进行分组，希望用户能指定分组的松紧程度，因此选择K-means聚类算法（可指定分组数量），但找不到相关示例，自己写的代码无法正确生成向量化句子——因为参考示例里用了TokenCountVectorizer中不存在的fitTransform方法。

原问题代码 #

require 'vendor/autoload.php';

use Phpml\Clustering\KMeans;
use Phpml\FeatureExtraction\TokenCountVectorizer;
use Phpml\Tokenization\WhitespaceTokenizer;

// Define the sentences
$sentences = [
    "This is the first sentence.",
    "The second sentence is here.",
    "Here is the third sentence.",
    "And the fourth sentence is here too.",
    "This is a different sentence.",
    "Another unique sentence here."
];

// Tokenize the sentences
$tokenizer = new WhitespaceTokenizer();
$tokenizedSentences = [];
foreach ($sentences as $sentence) {
    $tokens = $tokenizer->tokenize($sentence);
    $tokenizedSentences[] = $tokens;
}

// Vectorize the sentences
$vectorizer = new TokenCountVectorizer($tokenizer);
$vectorizedSentences = $vectorizer->fitTransform($tokenizedSentences);

// Perform K-means clustering
$kmeans = new KMeans(2);
$clusters = $kmeans->cluster($vectorizedSentences);

// Output the clusters
foreach ($clusters as $clusterId => $cluster) {
    echo "Cluster " . ($clusterId + 1) . ":\n";
    foreach ($cluster as $index) {
        echo "- " . $sentences[$index] . "\n";
    }
    echo "\n";
}

问题核心 #

TokenCountVectorizer类中并没有fitTransform方法，正确的流程是先调用fit()方法拟合语料构建词汇表，再调用transform()方法生成向量；同时手动分词的步骤是多余的，向量器会自动处理分词。

修复后的代码 #

require 'vendor/autoload.php';

use Phpml\Clustering\KMeans;
use Phpml\FeatureExtraction\TokenCountVectorizer;
use Phpml\Tokenization\WhitespaceTokenizer;

// 定义句子数组
$sentences = [
    "This is the first sentence.",
    "The second sentence is here.",
    "Here is the third sentence.",
    "And the fourth sentence is here too.",
    "This is a different sentence.",
    "Another unique sentence here."
];

// 初始化向量器，传入分词器
$vectorizer = new TokenCountVectorizer(new WhitespaceTokenizer());

// 拟合语料，构建词汇表
$vectorizer->fit($sentences);

// 将句子转换为词频向量
$vectorizedSentences = $vectorizer->transform($sentences);

// 执行K-means聚类，指定2个分组
$kmeans = new KMeans(2);
$clusters = $kmeans->cluster($vectorizedSentences);

// 输出聚类结果
foreach ($clusters as $clusterId => $cluster) {
    echo "Cluster " . ($clusterId + 1) . ":\n";
    foreach ($cluster as $index) {
        echo "- " . $sentences[$index] . "\n";
    }
    echo "\n";
}

关键修复点 #

• 移除手动分词步骤：TokenCountVectorizer会自动对输入的句子执行分词操作
• 拆分向量转换流程：
  • fit($sentences)：分析所有句子，提取并构建词汇表
  • transform($sentences)：基于词汇表，将每个句子转换为对应的词频向量

内容的提问来源于stack exchange，提问作者Mr J