gpt4 book ai didi

java - Lucene:Payloads 和相似性函数 --- 始终相同的 Payload 值

转载 作者:行者123 更新时间:2023-12-01 12:21:29 35 4
gpt4 key购买 nike

概述

我想实现一个 Lucene 索引器/搜索器,它使用新的有效负载功能,允许向文本添加元信息。在我的具体情况下,我向概念标签添加权重(可以理解为 % 概率,介于 0 到 100 之间),以便使用它们覆盖标准 Lucene TF-IDF 权重。我对此行为感到困惑,我相信我重写的相似性类有问题,但我无法弄清楚。

示例

当我运行搜索查询(例如“concept:red”)时,我发现每个有效负载始终是通过 MyPayloadSimilarity 传递的第一个数字(在代码示例中,这是 1.0),而不是 1.0、50.0 和 100.0。结果,所有文档都获得相同的有效负载和相同的分数。但是,数据应包含图片 #1,有效负载为 100.0,然后是图片 #2,然后是图片 #3 和非常不同的分数。我听不到我的声音。

以下是运行结果:

Query: concept:red
===> docid: 0 payload: 1.0
===> docid: 1 payload: 1.0
===> docid: 2 payload: 1.0
Number of results:3
-> docid: 3.jpg score: 0.2518424
-> docid: 2.jpg score: 0.2518424
-> docid: 1.jpg score: 0.2518424

出了什么问题?我对有效负载有什么误解吗?

代码

我将我的代码作为一个独立的示例分享,以便您尽可能轻松地运行它(如果您考虑此选项)。

public class PayloadShowcase {

public static void main(String s[]) {
PayloadShowcase p = new PayloadShowcase();
p.run();
}

public void run () {
// Step 1: indexing
MyPayloadIndexer indexer = new MyPayloadIndexer();
indexer.index();
// Step 2: searching
MyPayloadSearcher searcher = new MyPayloadSearcher();
searcher.search("red");
}

public class MyPayloadAnalyzer extends Analyzer {

private PayloadEncoder encoder;
MyPayloadAnalyzer(PayloadEncoder encoder) {
this.encoder = encoder;
}

@Override
protected TokenStreamComponents createComponents(String fieldName, Reader reader) {
Tokenizer source = new WhitespaceTokenizer(reader);
TokenStream filter = new LowerCaseFilter(source);
filter = new DelimitedPayloadTokenFilter(filter, '|', encoder);
return new TokenStreamComponents(source, filter);
}
}

public class MyPayloadIndexer {

public MyPayloadIndexer() {}

public void index() {
try {
Directory dir = FSDirectory.open(new File("D:/data/indices/sandbox"));
Analyzer analyzer = new MyPayloadAnalyzer(new FloatEncoder());
IndexWriterConfig iwconfig = new IndexWriterConfig(Version.LUCENE_4_10_1, analyzer);
iwconfig.setSimilarity(new MyPayloadSimilarity());
iwconfig.setOpenMode(IndexWriterConfig.OpenMode.CREATE);

// load mappings and classifiers
HashMap<String, String> mappings = this.loadDataMappings();
HashMap<String, HashMap> cMaps = this.loadData();

IndexWriter writer = new IndexWriter(dir, iwconfig);
indexDocuments(writer, mappings, cMaps);
writer.close();

} catch (IOException e) {
System.out.println("Exception while indexing: " + e.getMessage());
}
}

private void indexDocuments(IndexWriter writer, HashMap<String, String> fileMappings, HashMap<String, HashMap> concepts) throws IOException {

Set fileSet = fileMappings.keySet();
Iterator<String> iterator = fileSet.iterator();
while (iterator.hasNext()){
// unique file information
String fileID = iterator.next();
String filePath = fileMappings.get(fileID);
// create a new, empty document
Document doc = new Document();
// path of the indexed file
Field pathField = new StringField("path", filePath, Field.Store.YES);
doc.add(pathField);
// lookup all concept probabilities for this fileID
Iterator<String> conceptIterator = concepts.keySet().iterator();
while (conceptIterator.hasNext()){
String conceptName = conceptIterator.next();
HashMap conceptMap = concepts.get(conceptName);
doc.add(new TextField("concept", ("" + conceptName + "|").trim() + (conceptMap.get(fileID) + "").trim(), Field.Store.YES));
}
writer.addDocument(doc);
}
}

public HashMap<String, String> loadDataMappings(){
HashMap<String, String> h = new HashMap<>();
h.put("1", "1.jpg");
h.put("2", "2.jpg");
h.put("3", "3.jpg");
return h;
}

public HashMap<String, HashMap> loadData(){
HashMap<String, HashMap> h = new HashMap<>();
HashMap<String, String> green = new HashMap<>();
green.put("1", "50.0");
green.put("2", "1.0");
green.put("3", "100.0");
HashMap<String, String> red = new HashMap<>();
red.put("1", "100.0");
red.put("2", "50.0");
red.put("3", "1.0");
HashMap<String, String> blue = new HashMap<>();
blue.put("1", "1.0");
blue.put("2", "50.0");
blue.put("3", "100.0");
h.put("green", green);
h.put("red", red);
h.put("blue", blue);
return h;
}
}

class MyPayloadSimilarity extends DefaultSimilarity {

@Override
public float scorePayload(int docID, int start, int end, BytesRef payload) {
float pload = 1.0f;
if (payload != null) {
pload = PayloadHelper.decodeFloat(payload.bytes);
}
System.out.println("===> docid: " + docID + " payload: " + pload);
return pload;
}
}

public class MyPayloadSearcher {

public MyPayloadSearcher() {}

public void search(String queryString) {
try {
IndexReader reader = DirectoryReader.open(FSDirectory.open(new File("D:/data/indices/sandbox")));
IndexSearcher searcher = new IndexSearcher(reader);
searcher.setSimilarity(new PayloadSimilarity());
PayloadTermQuery query = new PayloadTermQuery(new Term("concept", queryString),
new AveragePayloadFunction());
System.out.println("Query: " + query.toString());
TopDocs topDocs = searcher.search(query, 999);
ScoreDoc[] hits = topDocs.scoreDocs;
System.out.println("Number of results:" + hits.length);

// output
for (int i = 0; i < hits.length; i++) {
Document doc = searcher.doc(hits[i].doc);
System.out.println("-> docid: " + doc.get("path") + " score: " + hits[i].score);
}
reader.close();

} catch (Exception e) {
System.out.println("Exception while searching: " + e.getMessage());
}
}
}

}

最佳答案

MyPayloadSimilarity 处,PayloadHelper.decodeFloat 调用不正确。在这种情况下,还需要传递 payload.offset 参数,如下所示:

pload = PayloadHelper.decodeFloat(payload.bytes, payload.offset);

希望对您有所帮助。

关于java - Lucene:Payloads 和相似性函数 --- 始终相同的 Payload 值,我们在Stack Overflow上找到一个类似的问题: https://stackoverflow.com/questions/26636284/

35 4 0
Copyright 2021 - 2024 cfsdn All Rights Reserved 蜀ICP备2022000587号
广告合作:1813099741@qq.com 6ren.com