使用deeplearning4j训练Doc2Vec(文档向量)

本文作者：合肥工业大学 管理学院 钱洋 email：1563178220@qq.com 内容可能有不到之处，欢迎交流。

未经本人允许禁止转载。

训练文档向量

在上一小节中，本人介绍了使用DeepLearning4J训练得到词向量(https://blog.csdn.net/qy20115549/article/details/82152462)。本篇主要介绍给定任意文本数据(分词后的数据)，如何使用DeepLearning4J训练得到文档的向量。

在平时的使用中，我们可以将文档转化成向量形式，进而进行聚类分类等其他操作。常用的将文档转化成向量形式的方法有one-hot编码、TF-IDF编码、主题模型(LDA)以及本篇要介绍的Doc2Vec操作。如下为笔者使用的文本数据：

对应的操作程序如下：

package org.deeplearning4j.examples.nlp.paragraphvectors;

import java.io.File;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collection;
import java.util.List;
import org.deeplearning4j.models.embeddings.loader.WordVectorSerializer;
import org.deeplearning4j.models.paragraphvectors.ParagraphVectors;
import org.deeplearning4j.models.word2vec.VocabWord;
import org.deeplearning4j.models.word2vec.wordstore.inmemory.AbstractCache;
import org.deeplearning4j.text.documentiterator.LabelsSource;
import org.deeplearning4j.text.sentenceiterator.LineSentenceIterator;
import org.deeplearning4j.text.sentenceiterator.SentenceIterator;
import org.deeplearning4j.text.tokenization.tokenizer.preprocessor.CommonPreprocessor;
import org.deeplearning4j.text.tokenization.tokenizerfactory.DefaultTokenizerFactory;
import org.deeplearning4j.text.tokenization.tokenizerfactory.TokenizerFactory;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;


public class Doc2VecTest {
    private static Logger log = LoggerFactory.getLogger(Doc2VecTest.class);
    //文档向量输出路径
    private static String outputPath = "E:/doc2vec.txt";
    public static void main(String[] args) throws Exception {
        //输入文本文件的目录
        File inputTxt = new File("E:/raw_sentences.txt");
        log.info("开始加载数据...."+inputTxt.getName());
        //加载数据
        SentenceIterator iter = new LineSentenceIterator(inputTxt);
        //切词操作
        TokenizerFactory token = new DefaultTokenizerFactory();
        //去除特殊符号及大小写转换操作
        token.setTokenPreProcessor(new CommonPreprocessor());
        AbstractCache cache=new AbstractCache<>();
        //添加文档标签，这个一般从文件读取，为了方面我这里使用了数字
        List labelList = new ArrayList();
        for (int i = 0; i < 97162; i++) {
            labelList.add("doc"+i);
        }
        //设置文档标签
        LabelsSource source = new LabelsSource(labelList);
        log.info("训练模型....");
        ParagraphVectors vec = new ParagraphVectors.Builder()
                .minWordFrequency(1)
                .iterations(5)
                .epochs(1)
                .layerSize(100)
                .learningRate(0.025)
                .labelsSource(source)
                .windowSize(5)
                .iterate(iter)
                .trainWordVectors(false)
                .vocabCache(cache)
                .tokenizerFactory(token)
                .sampling(0)
                .build();

        vec.fit();
        log.info("相似的句子:");
        Collection lst = vec.wordsNearest("doc0", 10);
        System.out.println(lst);
        log.info("输出文档向量....");
        WordVectorSerializer.writeWordVectors(vec, outputPath);
        //获取某词对应的向量
        log.info("向量获取:");
        double[] docVector = vec.getWordVector("doc0");
        System.out.println(Arrays.toString(docVector));
    }
}

程序在控制台输出的结果为：

另外，我们也成功的将每篇文档对应的向量输出到本地文件中，如下图所示为每篇文档对应的向量结果：

改写输出方式

另外，我们也可以写一个操作方法，是的输出结果按照每个人的需求来，比如我个人的需求是：

文档内容 Tab键分割 向量内容

那么，上述的程序可以重写为：

package com.qian;

import java.io.BufferedReader;
import java.io.BufferedWriter;
import java.io.File;
import java.io.FileInputStream;
import java.io.FileOutputStream;
import java.io.IOException;
import java.io.InputStreamReader;
import java.io.OutputStreamWriter;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collection;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import org.deeplearning4j.models.paragraphvectors.ParagraphVectors;
import org.deeplearning4j.models.word2vec.VocabWord;
import org.deeplearning4j.models.word2vec.wordstore.VocabCache;
import org.deeplearning4j.models.word2vec.wordstore.inmemory.AbstractCache;
import org.deeplearning4j.text.documentiterator.LabelsSource;
import org.deeplearning4j.text.sentenceiterator.LineSentenceIterator;
import org.deeplearning4j.text.sentenceiterator.SentenceIterator;
import org.deeplearning4j.text.tokenization.tokenizer.preprocessor.CommonPreprocessor;
import org.deeplearning4j.text.tokenization.tokenizerfactory.DefaultTokenizerFactory;
import org.deeplearning4j.text.tokenization.tokenizerfactory.TokenizerFactory;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;


public class Doc2VecTest {
    private static Logger log = LoggerFactory.getLogger(Doc2VecTest.class);
    //文档向量输出路径
    private static String outputPath = "data/doc2vec.txt";
    private static String inputPath = "data/raw_sentences.txt";
    public static void main(String[] args) throws Exception {
        //输入文本文件的目录
        File inputTxt = new File(inputPath);
        log.info("开始加载数据...." + inputTxt.getName());
        //加载数据
        SentenceIterator iter = new LineSentenceIterator(inputTxt);
        //切词操作
        TokenizerFactory token = new DefaultTokenizerFactory();
        //去除特殊符号及大小写转换操作
        token.setTokenPreProcessor(new CommonPreprocessor());
        AbstractCache cache=new AbstractCache<>();
        //添加文档标签，这个一般从文件读取，为了方面我这里使用了数字
        List labelList = new ArrayList();
        for (int i = 1; i < 97163; i++) {
            labelList.add("doc"+i);
        }
        //设置文档标签
        LabelsSource source = new LabelsSource(labelList);
        log.info("训练模型....");
        ParagraphVectors vec = new ParagraphVectors.Builder()
                .minWordFrequency(1)
                .iterations(5)
                .epochs(1)
                .layerSize(50)
                .learningRate(0.025)
                .labelsSource(source)
                .windowSize(5)
                .iterate(iter)
                .trainWordVectors(false)
                .vocabCache(cache)
                .tokenizerFactory(token)
                .sampling(0)
                .build();

        vec.fit();
        log.info("相似的句子:");
        Collection lst = vec.wordsNearest("doc1", 10);
        System.out.println(lst);
        log.info("输出文档向量....");
        writeDocVectors(vec,outputPath);
        //获取某词对应的向量
        log.info("向量获取:");
        double[] docVector = vec.getWordVector("doc1");
        System.out.println(Arrays.toString(docVector));
    }
    public static void writeDocVectors(ParagraphVectors vectors, String outpath) throws IOException {
        //写操作
        BufferedWriter bufferedWriter = new BufferedWriter(new OutputStreamWriter(new FileOutputStream(new File(outpath)),"gbk"));
        //读操作
        BufferedReader bufferedReader = new BufferedReader(new InputStreamReader(new FileInputStream(new File(inputPath)), "gbk")); 
        String line = null;
        int i = 1;
        Map keyToDoc = new HashMap<>();
        while ((line = bufferedReader.readLine())!=null) {
            keyToDoc.put("doc" + i, line);
            i++;
        }
        VocabCache vocabCache = vectors.getVocab();
        for (VocabWord word : vocabCache.vocabWords()) {
            StringBuilder builder = new StringBuilder();
            //获取每个文档对应的标签
            INDArray vector = vectors.getWordVectorMatrix(word.getLabel());
            //向量添加
            for (int j = 0; j < vector.length(); j++) {
                builder.append(vector.getDouble(j));
                if (j < vector.length() - 1) {
                    builder.append(" ");
                }
            }
            //写入指定文件
            bufferedWriter.write(keyToDoc.get(word.getLabel()) + "\t" + builder.append("\n").toString());
        }
        bufferedWriter.close();
        bufferedReader.close();
    }
}

程序的输出结果如下图所示，前面是文档后面是文档对应的向量：