Community Learn Tools Library Leisure
search
English
Home > Java > javaTutorial > Detailed explanation of how to use Java open source 11 Chinese word segmenters and comparison of word segmentation effects

Detailed explanation of how to use Java open source 11 Chinese word segmenters and comparison of word segmentation effects

黄舟
Release： 2017-03-22 11:00:47
Original
2148 people have browsed it

The goals of this article are twofold:

1. Learn to use the 11 major Java open source Chinese word segmenters

2. Comparatively analyze the word segmentation effects of the 11 major Java open source Chinese word segmenters

This article gives the usage methods of 11 major Java open source Chinese word segmentation and the comparison code of word segmentation results. As for which effect is better, the user should judge it by himself based on his own application scenarios.

11 Big Java open source Chinese word segmenters. Different word segmenters have different usages and different defined interfaces. Let’s first define a unified interface: 

/**
 * 获取文本的所有分词结果, 对比不同分词器结果
 * @author 杨尚川
 */
public interface WordSegmenter {
    /**
     * 获取文本的所有分词结果
     * @param text 文本
     * @return 所有的分词结果，去除重复
     */
    default public Set<String> seg(String text) {
        return segMore(text).values().stream().collect(Collectors.toSet());
    }
    /**
     * 获取文本的所有分词结果
     * @param text 文本
     * @return 所有的分词结果，KEY 为分词器模式，VALUE 为分词器结果
     */
    public Map<String, String> segMore(String text);
}
Copy after login

From the above definition we You know, in Java, the same method name and parameters, but different return values, overloading cannot be used in this case.

The difference between these two methods is the return value. Each word segmenter may have multiple word segmentation modes, and the word segmentation results of each mode may be different. The first method ignores the word segmentation mode and returns all All unique word segmentation results of the pattern. The second method returns each word segmentation mode and its corresponding word segmentation results.

Here, it should be noted that we use the new feature default method in Java8 and use stream to convert the value of a map into a unique collection.

Below we use these 11 major word segmenters to implement this interface:

1, word segmenter 

@Override
public Map<String, String> segMore(String text) {
    Map<String, String> map = new HashMap<>();
    for(SegmentationAlgorithm segmentationAlgorithm : SegmentationAlgorithm.values()){
        map.put(segmentationAlgorithm.getDes(), seg(text, segmentationAlgorithm));
    }
    return map;
}
private static String seg(String text, SegmentationAlgorithm segmentationAlgorithm) {
    StringBuilder result = new StringBuilder();
    for(Word word : WordSegmenter.segWithStopWords(text, segmentationAlgorithm)){
        result.append(word.getText()).append(" ");
    }
    return result.toString();
}
Copy after login

2, Ansj word segmenter 

@Override
public Map<String, String> segMore(String text) {
    Map<String, String> map = new HashMap<>();

    StringBuilder result = new StringBuilder();
    for(Term term : BaseAnalysis.parse(text)){
        result.append(term.getName()).append(" ");
    }
    map.put("BaseAnalysis", result.toString());

    result.setLength(0);
    for(Term term : ToAnalysis.parse(text)){
        result.append(term.getName()).append(" ");
    }
    map.put("ToAnalysis", result.toString());

    result.setLength(0);
    for(Term term : NlpAnalysis.parse(text)){
        result.append(term.getName()).append(" ");
    }
    map.put("NlpAnalysis", result.toString());

    result.setLength(0);
    for(Term term : IndexAnalysis.parse(text)){
        result.append(term.getName()).append(" ");
    }
    map.put("IndexAnalysis", result.toString());

    return map;
}
Copy after login

3 , Stanford tokenizer

private static final StanfordCoreNLP CTB = new StanfordCoreNLP("StanfordCoreNLP-chinese-ctb");
private static final StanfordCoreNLP PKU = new StanfordCoreNLP("StanfordCoreNLP-chinese-pku");
private static final PrintStream NULL_PRINT_STREAM = new PrintStream(new NullOutputStream(), false);
public Map<String, String> segMore(String text) {
    Map<String, String> map = new HashMap<>();
    map.put("Stanford Beijing University segmentation", seg(PKU, text));
    map.put("Stanford Chinese Treebank segmentation", seg(CTB, text));
    return map;
}
private static String seg(StanfordCoreNLP stanfordCoreNLP, String text){
    PrintStream err = System.err;
    System.setErr(NULL_PRINT_STREAM);
    Annotation document = new Annotation(text);
    stanfordCoreNLP.annotate(document);
    List<CoreMap> sentences = document.get(CoreAnnotations.SentencesAnnotation.class);
    StringBuilder result = new StringBuilder();
    for(CoreMap sentence: sentences) {
        for (CoreLabel token: sentence.get(CoreAnnotations.TokensAnnotation.class)) {
            String word = token.get(CoreAnnotations.TextAnnotation.class);;
            result.append(word).append(" ");
        }
    }
    System.setErr(err);
    return result.toString();
}
Copy after login

4, FudanNLP tokenizer

private static CWSTagger tagger = null;
static{
    try{
        tagger = new CWSTagger("lib/fudannlp_seg.m");
        tagger.setEnFilter(true);
    }catch(Exception e){
        e.printStackTrace();
    }
}
@Override
public Map<String, String> segMore(String text) {
    Map<String, String> map = new HashMap<>();
    map.put("FudanNLP", tagger.tag(text));
    return map;
}
Copy after login

5, Jieba tokenizer

private static final JiebaSegmenter JIEBA_SEGMENTER = new JiebaSegmenter();
@Override
public Map<String, String> segMore(String text) {
    Map<String, String> map = new HashMap<>();
    map.put("INDEX", seg(text, SegMode.INDEX));
    map.put("SEARCH", seg(text, SegMode.SEARCH));
    return map;
}
private static String seg(String text, SegMode segMode) {
    StringBuilder result = new StringBuilder();                
    for(SegToken token : JIEBA_SEGMENTER.process(text, segMode)){
        result.append(token.word.getToken()).append(" ");
    }
    return result.toString(); 
}
Copy after login

6, Jcseg tokenizer

private static final JcsegTaskConfig CONFIG = new JcsegTaskConfig();
private static final ADictionary DIC = DictionaryFactory.createDefaultDictionary(CONFIG);
static {
    CONFIG.setLoadCJKSyn(false);
    CONFIG.setLoadCJKPinyin(false);
}
@Override
public Map<String, String> segMore(String text) {
    Map<String, String> map = new HashMap<>();

    map.put("复杂模式", segText(text, JcsegTaskConfig.COMPLEX_MODE));
    map.put("简易模式", segText(text, JcsegTaskConfig.SIMPLE_MODE));

    return map;
}
private String segText(String text, int segMode) {
    StringBuilder result = new StringBuilder();        
    try {
        ISegment seg = SegmentFactory.createJcseg(segMode, new Object[]{new StringReader(text), CONFIG, DIC});
        IWord word = null;
        while((word=seg.next())!=null) {         
            result.append(word.getValue()).append(" ");
        }
    } catch (Exception ex) {
        throw new RuntimeException(ex);
    }
    return result.toString();
}
Copy after login

7, MMSeg4j Tokenizer

private static final Dictionary DIC = Dictionary.getInstance();
private static final SimpleSeg SIMPLE_SEG = new SimpleSeg(DIC);
private static final ComplexSeg COMPLEX_SEG = new ComplexSeg(DIC);
private static final MaxWordSeg MAX_WORD_SEG = new MaxWordSeg(DIC);
@Override
public Map<String, String> segMore(String text) {
    Map<String, String> map = new HashMap<>();
    map.put(SIMPLE_SEG.getClass().getSimpleName(), segText(text, SIMPLE_SEG));
    map.put(COMPLEX_SEG.getClass().getSimpleName(), segText(text, COMPLEX_SEG));
    map.put(MAX_WORD_SEG.getClass().getSimpleName(), segText(text, MAX_WORD_SEG));
    return map;
}
private String segText(String text, Seg seg) {
    StringBuilder result = new StringBuilder();
    MMSeg mmSeg = new MMSeg(new StringReader(text), seg);        
    try {
        Word word = null;
        while((word=mmSeg.next())!=null) {       
            result.append(word.getString()).append(" ");
        }
    } catch (IOException ex) {
        throw new RuntimeException(ex);
    }
    return result.toString();
}
Copy after login

8, IKAnalyzer tokenizer

@Override
public Map<String, String> segMore(String text) {
    Map<String, String> map = new HashMap<>();

    map.put("智能切分", segText(text, true));
    map.put("细粒度切分", segText(text, false));

    return map;
}
private String segText(String text, boolean useSmart) {
    StringBuilder result = new StringBuilder();
    IKSegmenter ik = new IKSegmenter(new StringReader(text), useSmart);        
    try {
        Lexeme word = null;
        while((word=ik.next())!=null) {          
            result.append(word.getLexemeText()).append(" ");
        }
    } catch (IOException ex) {
        throw new RuntimeException(ex);
    }
    return result.toString();
}
Copy after login

9, Paoding tokenizer

private static final PaodingAnalyzer ANALYZER = new PaodingAnalyzer();
@Override
public Map<String, String> segMore(String text) {
    Map<String, String> map = new HashMap<>();

    map.put("MOST_WORDS_MODE", seg(text, PaodingAnalyzer.MOST_WORDS_MODE));
    map.put("MAX_WORD_LENGTH_MODE", seg(text, PaodingAnalyzer.MAX_WORD_LENGTH_MODE));

    return map;
}
private static String seg(String text, int mode){
    ANALYZER.setMode(mode);
    StringBuilder result = new StringBuilder();
    try {
        Token reusableToken = new Token();
        TokenStream stream = ANALYZER.tokenStream("", new StringReader(text));
        Token token = null;
        while((token = stream.next(reusableToken)) != null){
            result.append(token.term()).append(" ");
        }
    } catch (Exception ex) {
        throw new RuntimeException(ex);
    }
    return result.toString();          
}
Copy after login

10, smartcn tokenizer

private static final SmartChineseAnalyzer SMART_CHINESE_ANALYZER = new SmartChineseAnalyzer();
@Override
public Map<String, String> segMore(String text) {
    Map<String, String> map = new HashMap<>();
    map.put("smartcn", segText(text));
    return map;
}
private static String segText(String text) {
    StringBuilder result = new StringBuilder();
    try {
        TokenStream tokenStream = SMART_CHINESE_ANALYZER.tokenStream("text", new StringReader(text));
        tokenStream.reset();
        while (tokenStream.incrementToken()){
            CharTermAttribute charTermAttribute = tokenStream.getAttribute(CharTermAttribute.class);
            result.append(charTermAttribute.toString()).append(" ");
        }
        tokenStream.close();
    }catch (Exception e){
        e.printStackTrace();
    }
    return result.toString();
}
Copy after login

11, HanLP tokenizer 

private static final Segment N_SHORT_SEGMENT = new NShortSegment().enableCustomDictionary(false).enablePlaceRecognize(true).enableOrganizationRecognize(true);
private static final Segment DIJKSTRA_SEGMENT = new DijkstraSegment().enableCustomDictionary(false).enablePlaceRecognize(true).enableOrganizationRecognize(true);
@Override
public Map<String, String> segMore(String text) {
    Map<String, String> map = new HashMap<>();
    map.put("标准分词", standard(text));
    map.put("NLP分词", nlp(text));
    map.put("索引分词", index(text));
    map.put("N-最短路径分词", nShort(text));
    map.put("最短路径分词", shortest(text));
    map.put("极速词典分词", speed(text));
    return map;
}
private static String standard(String text) {
    StringBuilder result = new StringBuilder();
    StandardTokenizer.segment(text).forEach(term->result.append(term.word).append(" "));
    return result.toString();
}
private static String nlp(String text) {
    StringBuilder result = new StringBuilder();
    NLPTokenizer.segment(text).forEach(term->result.append(term.word).append(" "));
    return result.toString();
}
private static String index(String text) {
    StringBuilder result = new StringBuilder();
    IndexTokenizer.segment(text).forEach(term->result.append(term.word).append(" "));
    return result.toString();
}
private static String speed(String text) {
    StringBuilder result = new StringBuilder();
    SpeedTokenizer.segment(text).forEach(term->result.append(term.word).append(" "));
    return result.toString();
}
private static String nShort(String text) {
    StringBuilder result = new StringBuilder();
    N_SHORT_SEGMENT.seg(text).forEach(term->result.append(term.word).append(" "));
    return result.toString();
}
private static String shortest(String text) {
    StringBuilder result = new StringBuilder();
    DIJKSTRA_SEGMENT.seg(text).forEach(term->result.append(term.word).append(" "));
    return result.toString();
}
Copy after login

Now we have achieved the first purpose of this article: learn to use the 11 major Java open source Chinese word segmenters.

Finally, let’s realize the second purpose of this article: comparative analysis of the word segmentation effects of the 11 major Java open source Chinese word segmenters. The program is as follows: 

public static Map<String, Set<String>> contrast(String text){
    Map<String, Set<String>> map = new LinkedHashMap<>();
    map.put("word分词器", new WordEvaluation().seg(text));
    map.put("Stanford分词器", new StanfordEvaluation().seg(text));
    map.put("Ansj分词器", new AnsjEvaluation().seg(text));
    map.put("HanLP分词器", new HanLPEvaluation().seg(text));
    map.put("FudanNLP分词器", new FudanNLPEvaluation().seg(text));
    map.put("Jieba分词器", new JiebaEvaluation().seg(text));
    map.put("Jcseg分词器", new JcsegEvaluation().seg(text));
    map.put("MMSeg4j分词器", new MMSeg4jEvaluation().seg(text));
    map.put("IKAnalyzer分词器", new IKAnalyzerEvaluation().seg(text));
    map.put("smartcn分词器", new SmartCNEvaluation().seg(text));
    return map;
}
public static Map<String, Map<String, String>> contrastMore(String text){
    Map<String, Map<String, String>> map = new LinkedHashMap<>();
    map.put("word分词器", new WordEvaluation().segMore(text));
    map.put("Stanford分词器", new StanfordEvaluation().segMore(text));
    map.put("Ansj分词器", new AnsjEvaluation().segMore(text));
    map.put("HanLP分词器", new HanLPEvaluation().segMore(text));
    map.put("FudanNLP分词器", new FudanNLPEvaluation().segMore(text));
    map.put("Jieba分词器", new JiebaEvaluation().segMore(text));
    map.put("Jcseg分词器", new JcsegEvaluation().segMore(text));
    map.put("MMSeg4j分词器", new MMSeg4jEvaluation().segMore(text));
    map.put("IKAnalyzer分词器", new IKAnalyzerEvaluation().segMore(text));
    map.put("smartcn分词器", new SmartCNEvaluation().segMore(text));
    return map;
}
public static void show(Map<String, Set<String>> map){
    map.keySet().forEach(k -> {
        System.out.println(k + " 的分词结果：");
        AtomicInteger i = new AtomicInteger();
        map.get(k).forEach(v -> {
            System.out.println("\t" + i.incrementAndGet() + " 、" + v);
        });
    });
}
public static void showMore(Map<String, Map<String, String>> map){
    map.keySet().forEach(k->{
        System.out.println(k + " 的分词结果：");
        AtomicInteger i = new AtomicInteger();
        map.get(k).keySet().forEach(a -> {
            System.out.println("\t" + i.incrementAndGet()+ " 、【"   + a + "】\t" + map.get(k).get(a));
        });
    });
}
public static void main(String[] args) {
    show(contrast("我爱楚离陌"));
    showMore(contrastMore("我爱楚离陌"));
}
Copy after login

The running results are as follows: 

********************************************
word分词器 的分词结果：
	1 、我 爱 楚离陌 
Stanford分词器 的分词结果：
	1 、我 爱 楚 离陌 
	2 、我 爱 楚离陌 
Ansj分词器 的分词结果：
	1 、我 爱 楚离 陌 
	2 、我 爱 楚 离 陌 
HanLP分词器 的分词结果：
	1 、我 爱 楚 离 陌 
smartcn分词器 的分词结果：
	1 、我 爱 楚 离 陌 
FudanNLP分词器 的分词结果：
	1 、我 爱楚离陌
Jieba分词器 的分词结果：
	1 、我爱楚 离 陌 
Jcseg分词器 的分词结果：
	1 、我 爱 楚 离 陌 
MMSeg4j分词器 的分词结果：
	1 、我爱 楚 离 陌 
IKAnalyzer分词器 的分词结果：
	1 、我 爱 楚 离 陌 
********************************************
Copy after login
********************************************
word分词器 的分词结果：
	1 、【全切分算法】	我 爱 楚离陌 
	2 、【双向最大最小匹配算法】	我 爱 楚离陌 
	3 、【正向最大匹配算法】	我 爱 楚离陌 
	4 、【双向最大匹配算法】	我 爱 楚离陌 
	5 、【逆向最大匹配算法】	我 爱 楚离陌 
	6 、【正向最小匹配算法】	我 爱 楚离陌 
	7 、【双向最小匹配算法】	我 爱 楚离陌 
	8 、【逆向最小匹配算法】	我 爱 楚离陌 
Stanford分词器 的分词结果：
	1 、【Stanford Chinese Treebank segmentation】	我 爱 楚离陌 
	2 、【Stanford Beijing University segmentation】	我 爱 楚 离陌 
Ansj分词器 的分词结果：
	1 、【BaseAnalysis】	我 爱 楚 离 陌 
	2 、【IndexAnalysis】	我 爱 楚 离 陌 
	3 、【ToAnalysis】	我 爱 楚 离 陌 
	4 、【NlpAnalysis】	我 爱 楚离 陌 
HanLP分词器 的分词结果：
	1 、【NLP分词】	我 爱 楚 离 陌 
	2 、【标准分词】	我 爱 楚 离 陌 
	3 、【N-最短路径分词】	我 爱 楚 离 陌 
	4 、【索引分词】	我 爱 楚 离 陌 
	5 、【最短路径分词】	我 爱 楚 离 陌 
	6 、【极速词典分词】	我 爱 楚 离 陌 
smartcn分词器 的分词结果：
	1 、【smartcn】	我 爱 楚 离 陌 
FudanNLP分词器 的分词结果：
	1 、【FudanNLP】	我 爱楚离陌
Jieba分词器 的分词结果：
	1 、【SEARCH】	我爱楚 离 陌 
	2 、【INDEX】	我爱楚 离 陌 
Jcseg分词器 的分词结果：
	1 、【简易模式】	我 爱 楚 离 陌 
	2 、【复杂模式】	我 爱 楚 离 陌 
MMSeg4j分词器 的分词结果：
	1 、【SimpleSeg】	我爱 楚 离 陌 
	2 、【ComplexSeg】	我爱 楚 离 陌 
	3 、【MaxWordSeg】	我爱 楚 离 陌 
IKAnalyzer分词器 的分词结果：
	1 、【智能切分】	我 爱 楚 离 陌 
	2 、【细粒度切分】	我 爱 楚 离 陌 
********************************************
Copy after login

The above is the detailed content of Detailed explanation of how to use Java open source 11 Chinese word segmenters and comparison of word segmentation effects. For more information, please follow other related articles on the PHP Chinese website!

Related labels：
Java，中文分词器
source：php.cn
Previous article：A detailed introduction to the core technical points of Java: the collection framework Next article：Detailed introduction to a comprehensive summary of Java container related knowledge (picture)
Statement of this Website
The content of this article is voluntarily contributed by netizens, and the copyright belongs to the original author. This site does not assume corresponding legal responsibility. If you find any content suspected of plagiarism or infringement, please contact admin@php.cn
Latest Articles by Author
Latest Issues
Custom hook (like) macros in React? I've only recently begun to understand what macros are in a Lisp context. My understanding...
From 2024-04-02 13:15:12
0
1
349
Use image sets to implement fallback background images without relying on javascript I'm fairly new to front-end and CSS. I'm trying to add a background-image and load .avif f...
From 2024-03-30 11:00:34
0
1
350
Related Topics
More>
Popular Recommendations
Popular Tutorials
More>
Related Tutorials
Popular Recommendations
Latest courses
Latest Downloads
More>
Web Effects
Website Source Code
Website Materials
Front End Template