Lucene排序 Payload的应用

有关Lucene的Payload的相关内容，可以参考如下链接，介绍的非常详细，值得参考：

http://www.ibm.com/developerworks/cn/opensource/os-cn-lucene-pl/
http://www.lucidimagination.com/blog/2009/08/05/getting-started-with-payloads/

例如，有这样的一个需求：

现在有两篇文档内容非常相似，如下所示：

文档1：egg tomato potato bread
文档2：egg book potato bread

现在我想要查询食物（foods），而且是查询关键词是egg，如何能够区别出上面两个文档哪一个更是我想要的？

可以看到上面两篇文档，文档1中描述的各项都是食物，而文档2中的book不是食物，基于上述需求，应该是文档1比文档2更相关，在查询结果中，文档1排名应该更靠前。通过上面
http://www.lucidimagination.com/blog/2009/08/05/getting-started-with-payloads/中给出的方法，可以在文档中，对给定词出现在文档的出现的权重信息（egg在文档1与文档中，以foods来衡量，文档1更相关），可以在索引之前处理一下，为egg增加payload信息，例如：

文档1：egg|0.984 tomato potato bread
文档2：egg|0.356 book potato bread

然后再进行索引，通过Lucene提供的PayloadTermQuery就能够分辨出上述egg这个Term的不同。在Lucene中，实际上是将我们存储的Payload数据，如上述"|"分隔后面的数字，乘到了tf上，然后在进行权重的计算。

下面，我们再看一下，增加一个Field来存储Payload数据，而源文档不需要进行修改，或者，我们可以在索引之前对文档进行一个处理，例如分类，通过分类可以给不同的文档所属类别的不同程度，计算一个Payload数值。

为了能够使用存储的Payload数据信息，结合上面提出的实例，我们需要按照如下步骤去做：

第一，待索引数据处理

例如，增加category这个Field存储类别信息，content这个Field存储上面的内容：

文档1：
new Field("category", "foods|0.984 shopping|0.503", Field.Store.YES, Field.Index.ANALYZED)
new Field("content", "egg tomato potato bread", Field.Store.YES, Field.Index.ANALYZED)
文档2：
new Field("category", "foods|0.356 shopping|0.791", Field.Store.YES, Field.Index.ANALYZED)
new Field("content", "egg book potato bread", Field.Store.YES, Field.Index.ANALYZED)

第二，实现解析Payload数据的Analyzer

由于Payload信息存储在category这个Field中，多个类别之间使用空格分隔，每个类别内容是以"|"分隔的，所以我们的Analyzer就要能够解析它。Lucene提供了DelimitedPayloadTokenFilter，能够处理具有分隔符的情况。我们的实现如下所示：

package org.shirdrn.lucene.query.payloadquery;

import java.io.Reader;

import org.apache.lucene.analysis.Analyzer;
import org.apache.lucene.analysis.TokenStream;
import org.apache.lucene.analysis.WhitespaceTokenizer;
import org.apache.lucene.analysis.payloads.DelimitedPayloadTokenFilter;
import org.apache.lucene.analysis.payloads.PayloadEncoder;

public class PayloadAnalyzer extends Analyzer {
private PayloadEncoder encoder;

PayloadAnalyzer(PayloadEncoder encoder) {
this.encoder = encoder;
}

@SuppressWarnings("deprecation")
public TokenStream tokenStream(String fieldName, Reader reader) {
TokenStream result = new WhitespaceTokenizer(reader); // 用来解析空格分隔的各个类别
result = new DelimitedPayloadTokenFilter(result, '|', encoder); // 在上面分词的基础上，在进行Payload数据解析
return result;
}
}

第三， 实现Similarity计算得分

Lucene中Similarity类中提供了scorePayload方法，用于计算Payload值来对文档贡献得分，我们重写了该方法，实现如下所示：

package org.shirdrn.lucene.query.payloadquery;

import org.apache.lucene.analysis.payloads.PayloadHelper;
import org.apache.lucene.search.DefaultSimilarity;

public class PayloadSimilarity extends DefaultSimilarity {

private static final long serialVersionUID = 1L;

@Override
public float scorePayload(int docId, String fieldName, int start, int end,
byte[] payload, int offset, int length) {
return PayloadHelper.decodeFloat(payload, offset);
}

}

通过使用PayloadHelper这个工具类可以获取到Payload值，然后在计算文档得分的时候起到作用。

第四，创建索引

在创建索引的时候，需要使用到我们上面实现的Analyzer和Similarity，代码如下所示：

package org.shirdrn.lucene.query.payloadquery;

import java.io.File;
import java.io.IOException;

import org.apache.lucene.analysis.Analyzer;
import org.apache.lucene.analysis.payloads.FloatEncoder;
import org.apache.lucene.document.Document;
import org.apache.lucene.document.Field;
import org.apache.lucene.index.CorruptIndexException;
import org.apache.lucene.index.IndexWriter;
import org.apache.lucene.index.IndexWriterConfig;
import org.apache.lucene.index.IndexWriterConfig.OpenMode;
import org.apache.lucene.search.Similarity;
import org.apache.lucene.store.FSDirectory;
import org.apache.lucene.store.LockObtainFailedException;
import org.apache.lucene.util.Version;

public class PayloadIndexing {

private IndexWriter indexWriter = null;
private final Analyzer analyzer = new PayloadAnalyzer(new FloatEncoder()); // 使用PayloadAnalyzer，并指定Encoder
private final Similarity similarity = new PayloadSimilarity(); // 实例化一个PayloadSimilarity
private IndexWriterConfig config = null;

public PayloadIndexing(String indexPath) throws CorruptIndexException, LockObtainFailedException, IOException {
File indexFile = new File(indexPath);
config = new IndexWriterConfig(Version.LUCENE_31, analyzer);
config.setOpenMode(OpenMode.CREATE).setSimilarity(similarity); // 设置计算得分的Similarity
indexWriter = new IndexWriter(FSDirectory.open(indexFile), config);
}

public void index() throws CorruptIndexException, IOException {
Document doc1 = new Document();
doc1.add(new Field("category", "foods|0.984 shopping|0.503", Field.Store.YES, Field.Index.ANALYZED));
doc1.add(new Field("content", "egg tomato potato bread", Field.Store.YES, Field.Index.ANALYZED));
indexWriter.addDocument(doc1);

Document doc2 = new Document();
doc2.add(new Field("category", "foods|0.356 shopping|0.791", Field.Store.YES, Field.Index.ANALYZED));
doc2.add(new Field("content", "egg book potato bread", Field.Store.YES, Field.Index.ANALYZED));
indexWriter.addDocument(doc2);

indexWriter.close();
}

public static void main(String[] args) throws CorruptIndexException, IOException {
new PayloadIndexing("E:\\index").index();
}
}

第五，查询

查询的时候，我们可以构造PayloadTermQuery来进行查询。代码如下所示：

package org.shirdrn.lucene.query.payloadquery;

import java.io.File;
import java.io.IOException;

import org.apache.lucene.document.Document;
import org.apache.lucene.index.CorruptIndexException;
import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.Term;
import org.apache.lucene.queryParser.ParseException;
import org.apache.lucene.search.BooleanQuery;
import org.apache.lucene.search.Explanation;
import org.apache.lucene.search.IndexSearcher;
import org.apache.lucene.search.ScoreDoc;
import org.apache.lucene.search.TopScoreDocCollector;
import org.apache.lucene.search.BooleanClause.Occur;
import org.apache.lucene.search.payloads.AveragePayloadFunction;
import org.apache.lucene.search.payloads.PayloadTermQuery;
import org.apache.lucene.store.NIOFSDirectory;

public class PayloadSearching {

private IndexReader indexReader;
private IndexSearcher searcher;

public PayloadSearching(String indexPath) throws CorruptIndexException, IOException {
indexReader = IndexReader.open(NIOFSDirectory.open(new File(indexPath)), true);
searcher = new IndexSearcher(indexReader);
searcher.setSimilarity(new PayloadSimilarity()); // 设置自定义的PayloadSimilarity
}

public ScoreDoc[] search(String qsr) throws ParseException, IOException {
int hitsPerPage = 10;
BooleanQuery bq = new BooleanQuery();
for(String q : qsr.split(" ")) {
bq.add(createPayloadTermQuery(q), Occur.MUST);
}
TopScoreDocCollector collector = TopScoreDocCollector.create(5 * hitsPerPage, true);
searcher.search(bq, collector);
ScoreDoc[] hits = collector.topDocs().scoreDocs;
for (int i = 0; i < hits.length; i++) {
int docId = hits[i].doc; // 文档编号
Explanation  explanation  = searcher.explain(bq, docId);
System.out.println(explanation.toString());
}
return hits;
}

public void display(ScoreDoc[] hits, int start, int end) throws CorruptIndexException, IOException {
end = Math.min(hits.length, end);
for (int i = start; i < end; i++) {
Document doc = searcher.doc(hits[i].doc);
int docId = hits[i].doc; // 文档编号
float score = hits[i].score; // 文档得分
System.out.println(docId + "\t" + score + "\t" + doc + "\t");
}
}

public void close() throws IOException {
searcher.close();
indexReader.close();
}

private PayloadTermQuery createPayloadTermQuery(String item) {
PayloadTermQuery ptq = null;
if(item.indexOf("^")!=-1) {
String[] a = item.split("\\^");
String field = a[0].split(":")[0];
String token = a[0].split(":")[1];
ptq = new PayloadTermQuery(new Term(field, token), new AveragePayloadFunction());
ptq.setBoost(Float.parseFloat(a[1].trim()));
} else {
String field = item.split(":")[0];
String token = item.split(":")[1];
ptq = new PayloadTermQuery(new Term(field, token), new AveragePayloadFunction());
}
return ptq;
}

public static void main(String[] args) throws ParseException, IOException {
int start = 0, end = 10;
//      String queries = "category:foods^123.0 content:bread^987.0";
String queries = "category:foods content:egg";
PayloadSearching payloadSearcher = new PayloadSearching("E:\\index");
payloadSearcher.display(payloadSearcher.search(queries), start, end);
payloadSearcher.close();
}

}

我们可以看到查询结果，两个文档的相关度排序：

0   0.3314532   Document<stored,indexed,tokenized<category:foods|0.984 shopping|0.503> stored,indexed,tokenized<content:egg tomato potato bread>>
1   0.21477573  Document<stored,indexed,tokenized<category:foods|0.356 shopping|0.791> stored,indexed,tokenized<content:egg book potato bread>>

通过输出计算得分的解释信息，如下所示：

0.3314532 = (MATCH) sum of:

0.18281947 = (MATCH) weight(category:foods in 0), product of:

0.70710677 = queryWeight(category:foods), product of:

0.5945349 = idf(category: foods=2)

1.1893445 = queryNorm

0.2585458 = (MATCH) fieldWeight(category:foods in 0), product of:

0.6957931 = (MATCH) btq, product of:

0.70710677 = tf(phraseFreq=0.5)

0.984 = scorePayload(...)

0.5945349 = idf(category: foods=2)

0.625 = fieldNorm(field=category, doc=0)

0.14863372 = (MATCH) weight(content:egg in 0), product of:

0.70710677 = queryWeight(content:egg), product of:

0.5945349 = idf(content: egg=2)

1.1893445 = queryNorm

0.21019982 = (MATCH) fieldWeight(content:egg in 0), product of:

0.70710677 = (MATCH) btq, product of:

0.70710677 = tf(phraseFreq=0.5)

1.0 = scorePayload(...)

0.5945349 = idf(content: egg=2)

0.5 = fieldNorm(field=content, doc=0)

0.21477571 = (MATCH) sum of:

0.066142 = (MATCH) weight(category:foods in 1), product of:

0.70710677 = queryWeight(category:foods), product of:

0.5945349 = idf(category: foods=2)

1.1893445 = queryNorm

0.09353892 = (MATCH) fieldWeight(category:foods in 1), product of:

0.25173002 = (MATCH) btq, product of:

0.70710677 = tf(phraseFreq=0.5)

0.356 = scorePayload(...)

0.5945349 = idf(category: foods=2)

0.625 = fieldNorm(field=category, doc=1)

0.14863372 = (MATCH) weight(content:egg in 1), product of:

0.70710677 = queryWeight(content:egg), product of:

0.5945349 = idf(content: egg=2)

1.1893445 = queryNorm

0.21019982 = (MATCH) fieldWeight(content:egg in 1), product of:

0.70710677 = (MATCH) btq, product of:

0.70710677 = tf(phraseFreq=0.5)

1.0 = scorePayload(...)

0.5945349 = idf(content: egg=2)

0.5 = fieldNorm(field=content, doc=1)

我们可以看到，除了在tf上乘了一个Payload值以外，其他的都相同，也就是说，我们预期使用的Payload为文档（ID=0）贡献了得分，排名靠前了。否则，如果不使用Payload的话，查询结果中两个文档的得分是相同的（可以模拟设置他们的Payload值相同，测试一下看看）

相关文章阅读及免费下载：

《Lucene Ranking算法分析》

《Lucene Payload 的研究与应用》

《Lucene排序 Payload的应用》

[b]《Apache Lucene3.0结果排序原理 操作 示例》[/b]

更多《Apache Lucene文档》，尽在开卷有益360 http://www.docin.com/book_360