Lucene文本解析器实现 把一段文本信息拆分成多个分词,我们都知道搜索引擎是通过分词检索的,文本解析器的好坏直接决定了搜索的精度和搜索的速度。
1.简单的Demo
private static final String[] examples = { "The quick brown 1234 fox jumped over the lazy dog!","XY&Z 15.6 Corporation - xyz@example.com", "北京市北京大学" }; private static final Analyzer[] ANALYZERS = new Analyzer[] {
new WhitespaceAnalyzer(), new SimpleAnalyzer(), new StopAnalyzer(), new StandardAnalyzer(), new CJKAnalyzer(), new SmartChineseAnalyzer() }; //空格符拆分 非字母拆分 非字母拆分去掉停词 Unicode文本分割 日韩文分割 简体中文分割
@Test public void testAnalyzer() throws IOException { for (int i = 0; i < ANALYZERS.length; i++) { String simpleName = ANALYZERS[i].getClass().getSimpleName(); for (int j = 0; j < examples.length; j++) { //TokenStream是分析处理组件中的一种中间数据格式,它从一个reader中获取文本, 分词器Tokenizer和过滤器TokenFilter继承自TokenStream TokenStream contents = ANALYZERS[i].tokenStream("contents", new StringReader(examples[j])); //添加多个Attribute,从而可以了解到分词之后详细的词元信息 ,OffsetAttribute 表示token的首字母和尾字母在原文本中的位置 OffsetAttribute offsetAttribute = contents.addAttribute(OffsetAttribute.class); TypeAttribute typeAttribute = contents.addAttribute(TypeAttribute.class); //TypeAttribute 表示token的词汇类型信息,默认值为word contents.reset(); System.out.println(" " + simpleName + " analyzing : " + examples[j]); while (contents.incrementToken()) { String s1 = offsetAttribute.toString(); int i1 = offsetAttribute.startOffset();// 起始偏移量 int i2 = offsetAttribute.endOffset(); // 结束偏移量 System.out.println(" " + s1 + "[" + i1 + "," + i2 + ":" + typeAttribute.type() + "]" + " "); } contents.end(); contents.close(); //调用incrementToken()结束迭代之后,调用end()和close()方法,其中end()可以唤醒当前TokenStream的处理器去做一些收尾工作,close()可以关闭TokenStream和Analyzer去释放在分析过程中使用的资源。 System.out.println(); } } } }
2. 了解tokenStream的Attribute
tokenStream()方法之后,添加多个Attribute,可以了解到分词之后详细的词元信息,比如CharTermAttribute用于保存词元的内容,TypeAttribute用于保存词元的类型。
CharTermAttribute 表示token本身的内容
PositionIncrementAttribute 表示当前token相对于前一个token的相对位置,也就是相隔的词语数量(例如“text for attribute”,
text和attribute之间的getPositionIncrement为2),如果两者之间没有停用词,那么该值被置为默认值1
OffsetAttribute 表示token的首字母和尾字母在原文本中的位置
TypeAttribute 表示token的词汇类型信息,默认值为word,
其它值有<ALPHANUM> <APOSTROPHE> <ACRONYM> <COMPANY> <EMAIL> <HOST> <NUM> <CJ> <ACRONYM_DEP>
FlagsAttribute 与TypeAttribute类似,假设你需要给token添加额外的信息,而且希望该信息可以通过分析链,那么就可以通过flags去传递
PayloadAttribute 在每个索引位置都存储了payload(关键信息),当使用基于Payload的查询时,该信息在评分中非常有用
@Test public void testAttribute() throws IOException { Analyzer analyzer = new StandardAnalyzer(); String input = "This is a test text for attribute! Just add-some word."; TokenStream tokenStream = analyzer.tokenStream("text", new StringReader(input)); CharTermAttribute charTermAttribute = tokenStream.addAttribute(CharTermAttribute.class); PositionIncrementAttribute positionIncrementAttribute = tokenStream.addAttribute(PositionIncrementAttribute.class); OffsetAttribute offsetAttribute = tokenStream.addAttribute(OffsetAttribute.class); TypeAttribute typeAttribute = tokenStream.addAttribute(TypeAttribute.class); PayloadAttribute payloadAttribute = tokenStream.addAttribute(PayloadAttribute.class); payloadAttribute.setPayload(new BytesRef("Just")); tokenStream.reset(); while (tokenStream.incrementToken()) { System.out.print( "[" + charTermAttribute + " increment:" + positionIncrementAttribute.getPositionIncrement() + " start:" + offsetAttribute.startOffset() + " end:" + offsetAttribute.endOffset() + " type:"+ typeAttribute.type() + " payload:" + payloadAttribute.getPayload() + "]\n"); } tokenStream.end(); tokenStream.close(); }
3.Lucene 的分词器Tokenizer和过滤器TokenFilter
一个分析器由一个分词器和多个过滤器组成,分词器接受reader数据转换成 TokenStream,TokenFilter主要用于TokenStream的过滤操作,用来处理Tokenizer或者上一个TokenFilter处理后的结果,如果是对现有分词器进行扩展或修改。
自定义TokenFilter需要实现incrementToken()抽象函数,
public class TestTokenFilter { @Test public void test() throws IOException { String text = "Hi, Dr Wang, Mr Liu asks if you stay with Mrs Liu yesterday!"; Analyzer analyzer = new WhitespaceAnalyzer(); CourtesyTitleFilter filter = new CourtesyTitleFilter(analyzer.tokenStream("text", text)); CharTermAttribute charTermAttribute = filter.addAttribute(CharTermAttribute.class); filter.reset(); while (filter.incrementToken()) { System.out.print(charTermAttribute + " "); } } } /** * 自定义词扩展过滤器 */ class CourtesyTitleFilter extends TokenFilter { Map<String, String> courtesyTitleMap = new HashMap<>(); private CharTermAttribute termAttribute; protected CourtesyTitleFilter(TokenStream input) { super(input); termAttribute = addAttribute(CharTermAttribute.class); courtesyTitleMap.put("Dr", "doctor"); courtesyTitleMap.put("Mr", "mister"); courtesyTitleMap.put("Mrs", "miss"); } @Override public final boolean incrementToken() throws IOException { if (!input.incrementToken()) { return false; } String small = termAttribute.toString(); if (courtesyTitleMap.containsKey(small)) { termAttribute.setEmpty().append(courtesyTitleMap.get(small)); } return true; } }
输出结果如下
Hi, doctor Wang, mister Liu asks if you stay with miss Liu yesterday!
4.自定义Analyzer实现扩展停用词
class StopAnalyzerExtend extends Analyzer { private CharArraySet stopWordSet;//停止词词典 public CharArraySet getStopWordSet() { return this.stopWordSet; } public void setStopWordSet(CharArraySet stopWordSet) { this.stopWordSet = stopWordSet; } public StopAnalyzerExtend() { super(); setStopWordSet(StopAnalyzer.ENGLISH_STOP_WORDS_SET); } /** * @param stops 需要扩展的停止词 */ public StopAnalyzerExtend(List<String> stops) { this(); /**如果直接为stopWordSet赋值的话,会报如下异常,这是因为在StopAnalyzer中有ENGLISH_STOP_WORDS_SET = CharArraySet.unmodifiableSet(stopSet); * ENGLISH_STOP_WORDS_SET 被设置为不可更改的set集合 */ //stopWordSet = getStopWordSet(); stopWordSet = CharArraySet.copy(getStopWordSet()); stopWordSet.addAll(StopFilter.makeStopSet(stops)); } @Override protected TokenStreamComponents createComponents(String fieldName) { Tokenizer source = new LowerCaseTokenizer(); return new TokenStreamComponents(source, new StopFilter(source, stopWordSet)); } public static void main(String[] args) throws IOException { ArrayList<String> strings = new ArrayList<String>() {{ add("小鬼子"); add("美国佬"); }}; Analyzer analyzer = new StopAnalyzerExtend(strings); String content = "小鬼子 and 美国佬 are playing together!"; TokenStream tokenStream = analyzer.tokenStream("myfield", content); tokenStream.reset(); CharTermAttribute charTermAttribute = tokenStream.addAttribute(CharTermAttribute.class); while (tokenStream.incrementToken()) { // 已经过滤掉自定义停用词 // 输出:playing together System.out.println(charTermAttribute.toString()); } tokenStream.end(); tokenStream.close(); } }
5.自定义Analyzer实现字长过滤
class LongFilterAnalyzer extends Analyzer { private int len; public int getLen() { return this.len; } public void setLen(int len) { this.len = len; } public LongFilterAnalyzer() { super(); } public LongFilterAnalyzer(int len) { super(); setLen(len); } @Override protected TokenStreamComponents createComponents(String fieldName) { final Tokenizer source = new WhitespaceTokenizer(); //过滤掉长度<len,并且>20的token TokenStream tokenStream = new LengthFilter(source, len, 20); return new TokenStreamComponents(source, tokenStream); } public static void main(String[] args) { //把长度小于2的过滤掉,开区间 Analyzer analyzer = new LongFilterAnalyzer(2); String words = "I am a java coder! Testingtestingtesting!"; TokenStream stream = analyzer.tokenStream("myfield", words); try { stream.reset(); CharTermAttribute offsetAtt = stream.addAttribute(CharTermAttribute.class); while (stream.incrementToken()) { System.out.println(offsetAtt.toString()); } stream.end(); stream.close(); } catch (IOException e) { } } }
长度小于两个字符的文本都被过滤掉了。
6.PerFieldAnalyzerWrapper 处理不同的Field使用不同的Analyzer 。PerFieldAnalyzerWrapper可以像其它的Analyzer一样使用,包括索引和查询分析
@Test public void testPerFieldAnalyzerWrapper() throws IOException, ParseException { Map<String, Analyzer> fields = new HashMap<>(); fields.put("partnum", new KeywordAnalyzer()); // 对于其他的域,默认使用SimpleAnalyzer分析器,对于指定的域partnum使用KeywordAnalyzer PerFieldAnalyzerWrapper perFieldAnalyzerWrapper = new PerFieldAnalyzerWrapper(new SimpleAnalyzer(), fields); Directory directory = new RAMDirectory(); IndexWriterConfig indexWriterConfig = new IndexWriterConfig(perFieldAnalyzerWrapper); IndexWriter indexWriter = new IndexWriter(directory, indexWriterConfig); Document document = new Document(); FieldType fieldType = new FieldType(); fieldType.setStored(true); fieldType.setIndexOptions(IndexOptions.DOCS_AND_FREQS); document.add(new Field("partnum", "Q36", fieldType)); document.add(new Field("description", "Illidium Space Modulator", fieldType)); indexWriter.addDocument(document); indexWriter.close(); IndexSearcher indexSearcher = new IndexSearcher(DirectoryReader.open(directory)); // 直接使用TermQuery是可以检索到的 TopDocs search = indexSearcher.search(new TermQuery(new Term("partnum", "Q36")), 10); Assert.assertEquals(1, search.totalHits); // 如果使用QueryParser,那么必须要使用PerFieldAnalyzerWrapper,否则如下所示,是检索不到的 Query description = new QueryParser("description", new SimpleAnalyzer()).parse("partnum:Q36 AND SPACE"); search = indexSearcher.search(description, 10); Assert.assertEquals(0, search.totalHits); System.out.println("SimpleAnalyzer :" + description.toString());// +partnum:q // +description:space,原因是SimpleAnalyzer会剥离非字母字符并将字母小写化 // 使用PerFieldAnalyzerWrapper可以检索到 // partnum:Q36 AND SPACE表示在partnum中出现Q36,在description中出现SPACE description = new QueryParser("description", perFieldAnalyzerWrapper).parse("partnum:Q36 AND SPACE"); search = indexSearcher.search(description, 10); Assert.assertEquals(1, search.totalHits); System.out.println("(SimpleAnalyzer,KeywordAnalyzer) :" + description.toString());// +partnum:Q36 +description:space }
参考 : http://www.codepub.cn/2016/05/23/Lucene-6-0-in-action-4-The-text-analyzer/