Lucene5学习之自定义Collector

        你们都睡了，而我却在写博客，呵呵！我也不知道为什么都夜深了，我却还没一点困意，趁着劲头赶紧把自定义结果集写完，已经拖了2天没更新了，不能让你们等太久，我也要把写博客一直坚持下去。

        Collector是什么？还是看源码吧。这也是最权威的解释说明。

/**
 * <p>Expert: Collectors are primarily meant to be used to
 * gather raw results from a search, and implement sorting
 * or custom result filtering, collation, etc. </p>
 *
 * <p>Lucene's core collectors are derived from {@link Collector}
 * and {@link SimpleCollector}. Likely your application can
 * use one of these classes, or subclass {@link TopDocsCollector},
 * instead of implementing Collector directly:
 *
 * <ul>
 *
 *   <li>{@link TopDocsCollector} is an abstract base class
 *   that assumes you will retrieve the top N docs,
 *   according to some criteria, after collection is
 *   done.  </li>
 *
 *   <li>{@link TopScoreDocCollector} is a concrete subclass
 *   {@link TopDocsCollector} and sorts according to score +
 *   docID.  This is used internally by the {@link
 *   IndexSearcher} search methods that do not take an
 *   explicit {@link Sort}. It is likely the most frequently
 *   used collector.</li>
 *
 *   <li>{@link TopFieldCollector} subclasses {@link
 *   TopDocsCollector} and sorts according to a specified
 *   {@link Sort} object (sort by field).  This is used
 *   internally by the {@link IndexSearcher} search methods
 *   that take an explicit {@link Sort}.
 *
 *   <li>{@link TimeLimitingCollector}, which wraps any other
 *   Collector and aborts the search if it's taken too much
 *   time.</li>
 *
 *   <li>{@link PositiveScoresOnlyCollector} wraps any other
 *   Collector and prevents collection of hits whose score
 *   is &lt;= 0.0</li>
 *
 * </ul>
 *
 * @lucene.experimental
 */
public interface Collector {

  /**
   * Create a new {@link LeafCollector collector} to collect the given context.
   *
   * @param context
   *          next atomic reader context
   */
  LeafCollector getLeafCollector(LeafReaderContext context) throws IOException;

}

    Collector系列接口是用来收集查询结果，实现排序，自定义结果集过滤和收集。Collector和LeafCollector是Lucene结果集收集的核心。

    TopDocsCollector：是用来收集Top N结果的，

    TopScoreDocCollector：它是TopDocsCollector的子类，它返回的结果集会根据评分和docId进行排序，该接口在IndexSearcher类的search方法内部被调用，但search方法并不需要显式的指定一个Sort排序器，TopScoreDocCollector是使用频率最高的一个结果收集器接口。

     TopFieldCollector：它也是TopDocsCollector的子类，跟TopScoreDocCollector的区别是，TopScoreDocCollector是根据评分和docId进行排序的，而TopFieldCollector是根据用户指定的域进行排序，在调用IndexSearcher.search方法时需要显式的指定Sort排序器。

      TimeLimitingCollector：它是其他Collector的包装器，它的功能是当被包装的Collector耗时超过限制时可以中断收集过程。

      PositiveScoresOnlyCollector：从类名就知道它是干嘛的，Positive正数的意思，即只返回score评分大于零的索引文档，它跟TimeLimitingCollector都属于其他Collector的包装器，都使用了装饰者模式。

 

      Collector接口只有一个接口方法：

LeafCollector getLeafCollector(LeafReaderContext context) throws IOException;

     根据提供的IndexReader上下文对象返回一个LeafCollector，LeafCollector其实就是对应每个段文件的收集器，每次切换段文件时都会调用一次此接口方法。

    其实LeafCollector才是结果收集器接口，Collector只是用来生成每个段文件对应的LeafCollector，在Lucene4,x时代，Collector和LeafCollector并没有分开，现在Lucene5.0中，接口定义粒度更细了，为用户自定义扩展提供了更多的便利。

    接着看看LeafCollector的源码说明：

/**
 * <p>Collector decouples the score from the collected doc:
 * the score computation is skipped entirely if it's not
 * needed.  Collectors that do need the score should
 * implement the {@link #setScorer} method, to hold onto the
 * passed {@link Scorer} instance, and call {@link
 * Scorer#score()} within the collect method to compute the
 * current hit's score.  If your collector may request the
 * score for a single hit multiple times, you should use
 * {@link ScoreCachingWrappingScorer}. </p>
 * 
 * <p><b>NOTE:</b> The doc that is passed to the collect
 * method is relative to the current reader. If your
 * collector needs to resolve this to the docID space of the
 * Multi*Reader, you must re-base it by recording the
 * docBase from the most recent setNextReader call.  Here's
 * a simple example showing how to collect docIDs into a
 * BitSet:</p>
 * 
 * <pre class="prettyprint">
 * IndexSearcher searcher = new IndexSearcher(indexReader);
 * final BitSet bits = new BitSet(indexReader.maxDoc());
 * searcher.search(query, new Collector() {
 *
 *   public LeafCollector getLeafCollector(LeafReaderContext context)
 *       throws IOException {
 *     final int docBase = context.docBase;
 *     return new LeafCollector() {
 *
 *       <em>// ignore scorer</em>
 *       public void setScorer(Scorer scorer) throws IOException {
 *       }
 *
 *       public void collect(int doc) throws IOException {
 *         bits.set(docBase + doc);
 *       }
 *
 *     };
 *   }
 *
 * });
 * </pre>
 *
 * <p>Not all collectors will need to rebase the docID.  For
 * example, a collector that simply counts the total number
 * of hits would skip it.</p>
 *
 * @lucene.experimental
 */
public interface LeafCollector {

  /**
   * Called before successive calls to {@link #collect(int)}. Implementations
   * that need the score of the current document (passed-in to
   * {@link #collect(int)}), should save the passed-in Scorer and call
   * scorer.score() when needed.
   */
  void setScorer(Scorer scorer) throws IOException;
  
  /**
   * Called once for every document matching a query, with the unbased document
   * number.
   * <p>Note: The collection of the current segment can be terminated by throwing
   * a {@link CollectionTerminatedException}. In this case, the last docs of the
   * current {@link org.apache.lucene.index.LeafReaderContext} will be skipped and {@link IndexSearcher}
   * will swallow the exception and continue collection with the next leaf.
   * <p>
   * Note: This is called in an inner search loop. For good search performance,
   * implementations of this method should not call {@link IndexSearcher#doc(int)} or
   * {@link org.apache.lucene.index.IndexReader#document(int)} on every hit.
   * Doing so can slow searches by an order of magnitude or more.
   */
  void collect(int doc) throws IOException;

}

    LeafCollector将打分操作从文档收集中分离出去了，如果你不需要打分操作，你可以完全跳过。

 如果你需要打分操作，你需要实现setScorer方法并传入一个Scorer对象，然后在collect方法中

 通过调用Scorer.score方法完成对当前命中文档的打分操作。如果你的LeafCollector在collect

 方法中需要对命中的某个索引文档调用多次score方法的话，请你使用ScoreCachingWrappingScorer

 对象包装你的Scorer对象。(利用缓存防止多次进行重复打分)

 collect方法中的doc参数是相对于当前IndexReader的，如果你需要把doc解析成docId(索引文档ID),

 你需要调用setNextReader方法来重新计算IndexReader的docBase值。

 并不是所有的Collector都需要计算docID基数的，比如对于只需要收集总的命中结果数量的Collector来说，

 可以跳过这个操作。

 

       通过以上的理解，我们可以总结出：通过Collector接口生产LeafCollector，然后通过LeafCollector接口

去完成结果收集和命中结果的打分操作。即底下真正干活的是LeafCollector。

void collect(int doc) throws IOException;

    这里collect方法用来收集每个索引文档，提供的doc参数表示段文件编号，如果你要获取索引文档的编号，请加上当前段文件Reader的docBase基数，如leafReaderContext.reader().docBase + doc;

    如果你需要自定义打分器，请继承实现自己的Scorer，那这个setScorer什么时候调用呢，这个通过阅读IndexSearcher的search方法顺藤摸瓜从而知晓，看图：

      其实内部是先把Query对象包装成Filter，然后通过调用createNormalizedWeight方法生成Weight(权重类)，观摩Weight接口你会发现，其中有个Scorer scorer接口方法：

      至此我们就弄清楚了，我们的LeafCollector不用关心Scorer是怎么创建并传入到LeafCollector中的，我们只需要实现自己的Scorer即可，我们在IndexSearcher.search方法时内部会首先创建Weight,通过Weight来生成Scorer，我们在调用search方法时需要传入collector接口，那自然scorer接口就被传入了leafCollector中。

      如果实现了自己的Scorer则必然需要也要实现自己的Weight并通过自定义Weight来生成自定义Scorer，特此提醒，为了简便起见，这里就没有自定义Scorer。

     下面是一个自定义Collector的简单示例，希望能抛砖引玉，为大家排忧解惑，如果代码有任何BUG或纰漏，还望大家告知我。

package com.yida.framework.lucene5.collector;

import java.io.IOException;
import java.util.ArrayList;
import java.util.List;

import org.apache.lucene.index.LeafReaderContext;
import org.apache.lucene.index.SortedDocValues;
import org.apache.lucene.search.Collector;
import org.apache.lucene.search.LeafCollector;
import org.apache.lucene.search.ScoreDoc;
import org.apache.lucene.search.Scorer;
/**
 * 自定义Collector结果收集器
 * @author Lanxiaowei
 *
 */
public class GroupCollector implements Collector, LeafCollector {
	/**评分计算器*/
	private Scorer scorer;
	/**段文件的编号*/
    private int docBase;
    
    private String fieldName;
    private SortedDocValues sortedDocValues;
    
    private List<ScoreDoc> scoreDocs = new ArrayList<ScoreDoc>();
    
    public LeafCollector getLeafCollector(LeafReaderContext context)
			throws IOException {
    	this.sortedDocValues = context.reader().getSortedDocValues(fieldName);
    	return this;
	}
    
	public void setScorer(Scorer scorer) throws IOException {
		this.scorer = scorer;
	}

	public void collect(int doc) throws IOException {
        // scoreDoc:docId和评分
        this.scoreDocs.add(new ScoreDoc(this.docBase + doc, this.scorer.score()));
	}

	public GroupCollector(String fieldName) {
		super();
		this.fieldName = fieldName;
	}

	public int getDocBase() {
		return docBase;
	}

	public void setDocBase(int docBase) {
		this.docBase = docBase;
	}

	public String getFieldName() {
		return fieldName;
	}

	public void setFieldName(String fieldName) {
		this.fieldName = fieldName;
	}

	public SortedDocValues getSortedDocValues() {
		return sortedDocValues;
	}

	public void setSortedDocValues(SortedDocValues sortedDocValues) {
		this.sortedDocValues = sortedDocValues;
	}

	public List<ScoreDoc> getScoreDocs() {
		return scoreDocs;
	}

	public void setScoreDocs(List<ScoreDoc> scoreDocs) {
		this.scoreDocs = scoreDocs;
	}

	public Scorer getScorer() {
		return scorer;
	}
}

    

package com.yida.framework.lucene5.collector;

import java.io.IOException;
import java.nio.file.Paths;
import java.util.List;

import org.apache.lucene.document.Document;
import org.apache.lucene.index.DirectoryReader;
import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.Term;
import org.apache.lucene.search.IndexSearcher;
import org.apache.lucene.search.ScoreDoc;
import org.apache.lucene.search.TermQuery;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.FSDirectory;
/**
 * 自定义Collector测试
 * @author Lanxiaowei
 *
 */
public class GroupCollectorTest {
	public static void main(String[] args) throws IOException {
		String indexDir = "C:/lucenedir";
		Directory directory = FSDirectory.open(Paths.get(indexDir));
	    IndexReader reader = DirectoryReader.open(directory);
	    IndexSearcher searcher = new IndexSearcher(reader);
	    TermQuery termQuery = new TermQuery(new Term("title", "lucene"));
	    GroupCollector collector = new GroupCollector("title2");
	    searcher.search(termQuery, null, collector);
	    List<ScoreDoc> docs = collector.getScoreDocs();
		for (ScoreDoc scoreDoc : docs) {
			int docID = scoreDoc.doc;
			Document document = searcher.doc(docID);
			String title = document.get("title");
			float score = scoreDoc.score;
			System.out.println(docID + ":" + title + "  " + score);
		}
	    
	    reader.close();
	    directory.close();
	}
}

    这里仅仅是一个简单的示例，如果你需要更严格的干预索引文档，请在collect方法里实现的代码逻辑，如果你需要更细粒度的干预文档打分过程，请继承Scorer抽象类自定义的实现并继承Weight抽象类自定义的实现，然后调用IndexSearch的这个方法即可：

protected TopFieldDocs search(Weight weight, FieldDoc after, int nDocs,
                                Sort sort, boolean fillFields,
                                boolean doDocScores, boolean doMaxScore)
      throws IOException

 

     一如既往的，demo源码会上传到底下的附件里，至于有童鞋要求我的demo不要使用Maven构建，I am very sorry,I can't meet your requirments.如果你不会Maven，还是花时间去学下吧。OK，凌晨一点多了，我该搁笔就寝咯！

       哥的QQ: 7-3-6-0-3-1-3-0-5,欢迎加入哥的Java技术群一起交流学习。

    群号： 

 

评论

1 楼 kimibaby1990 2017-08-23  

谢谢分享，学到很多东西~
有两处觉的有点小问题：
1.
>  你需要调用setNextReader方法来重新计算IndexReader的docBase值。
lucene5里setNextReader 好像改成了doSetNextReader().

2.
>  如果你要获取索引文档的编号，请加上当前段文件Reader的docBase基数，如
>  leafReaderContext.reader().docBase + doc;
lucene5里 docBase直接通过LeafReaderContext的成员docBase就能得到。