feat: 下沉知识库检索编排能力

- 新增 rag retrieval 核心协议、RRF 融合与相关度归一化

- 支持关键词检索按 knowledgeId 过滤并补充 ES/Lucene 单测

- 扩展 KnowledgeNode 检索模式与 Milvus 检索参数透传
This commit is contained in:
2026-04-05 20:22:59 +08:00
parent 941995d1b8
commit f57544daa2
28 changed files with 1309 additions and 34 deletions

View File

@@ -37,6 +37,11 @@
<artifactId>jackson-databind</artifactId>
<version>2.15.2</version> <!-- 或与Elasticsearch客户端兼容的版本 -->
</dependency>
<dependency>
<groupId>junit</groupId>
<artifactId>junit</artifactId>
<scope>test</scope>
</dependency>
</dependencies>
</project>

View File

@@ -4,12 +4,15 @@ import co.elastic.clients.elasticsearch.ElasticsearchClient;
import co.elastic.clients.elasticsearch.core.*;
import co.elastic.clients.elasticsearch.core.bulk.BulkOperation;
import co.elastic.clients.elasticsearch.core.bulk.IndexOperation;
import co.elastic.clients.elasticsearch.core.search.SourceConfig;
import co.elastic.clients.json.JsonData;
import co.elastic.clients.json.jackson.JacksonJsonpMapper;
import co.elastic.clients.transport.ElasticsearchTransport;
import co.elastic.clients.transport.rest_client.RestClientTransport;
import com.easyagents.core.document.Document;
import com.easyagents.search.engine.service.DocumentSearcher;
import com.easyagents.search.engine.service.KeywordSearchMetadataKeys;
import com.easyagents.search.engine.service.KeywordSearchRequest;
import org.apache.http.HttpHost;
import org.apache.http.auth.AuthScope;
import org.apache.http.auth.UsernamePasswordCredentials;
@@ -88,13 +91,7 @@ public class ElasticSearcher implements DocumentSearcher {
transport = new RestClientTransport(restClient, new JacksonJsonpMapper());
ElasticsearchClient client = new ElasticsearchClient(transport);
Map<String, Object> source = new HashMap<>();
source.put("id", document.getId());
source.put("content", document.getContent());
if (document.getTitle() != null) {
source.put("title", document.getTitle());
}
Map<String, Object> source = buildSource(document);
String documentId = document.getId().toString();
IndexOperation<?> indexOp = IndexOperation.of(i -> i
.index(esConfig.getIndexName())
@@ -116,7 +113,7 @@ public class ElasticSearcher implements DocumentSearcher {
}
@Override
public List<Document> searchDocuments(String keyword, int count) {
public List<Document> searchDocuments(KeywordSearchRequest request) {
RestClient restClient = null;
ElasticsearchTransport transport = null;
@@ -125,21 +122,16 @@ public class ElasticSearcher implements DocumentSearcher {
transport = new RestClientTransport(restClient, new JacksonJsonpMapper());
ElasticsearchClient client = new ElasticsearchClient(transport);
SearchRequest request = SearchRequest.of(s -> s
.index(esConfig.getIndexName())
.size(count)
.query(q -> q
.match(m -> m
.field("title")
.field("content")
.query(keyword)
)
)
);
SearchResponse<Document> response = client.search(request, Document.class);
SearchResponse<Map> response = client.search(buildSearchRequest(request), Map.class);
List<Document> results = new ArrayList<>();
response.hits().hits().forEach(hit -> results.add(hit.source()));
response.hits().hits().forEach(hit -> {
Map source = hit.source();
Document document = toDocument(hit.id(), source, hit.score());
if (document == null) {
return;
}
results.add(document);
});
return results;
} catch (Exception e) {
@@ -193,14 +185,17 @@ public class ElasticSearcher implements DocumentSearcher {
transport = new RestClientTransport(restClient, new JacksonJsonpMapper());
ElasticsearchClient client = new ElasticsearchClient(transport);
UpdateRequest<Document, Object> request = UpdateRequest.of(u -> u
UpdateRequest<Map<String, Object>, Map<String, Object>> request = UpdateRequest.of(u -> u
.index(esConfig.getIndexName())
.id(document.getId().toString())
.doc(document)
.doc(buildSource(document))
);
UpdateResponse<Document> response = client.update(request, Object.class);
return response.result() == co.elastic.clients.elasticsearch._types.Result.Updated;
@SuppressWarnings("unchecked")
Class<Map<String, Object>> documentClass = (Class<Map<String, Object>>) (Class<?>) Map.class;
UpdateResponse<Map<String, Object>> response = client.update(request, documentClass);
return response.result() == co.elastic.clients.elasticsearch._types.Result.Updated
|| response.result() == co.elastic.clients.elasticsearch._types.Result.NoOp;
} catch (Exception e) {
LOG.error("Error updating document with id: " + document.getId(), e);
return false;
@@ -220,4 +215,88 @@ public class ElasticSearcher implements DocumentSearcher {
}
}
}
@SuppressWarnings("unchecked")
private Document toDocument(String hitId, Map source, Double score) {
if (source == null || source.isEmpty()) {
return null;
}
Document document = new Document();
Object id = source.get("id");
document.setId(id != null ? id : hitId);
Object title = source.get("title");
if (title != null) {
document.setTitle(String.valueOf(title));
}
Object content = source.get("content");
if (content != null) {
document.setContent(String.valueOf(content));
}
Object metadataMap = source.get("metadataMap");
if (metadataMap instanceof Map<?, ?>) {
document.setMetadataMap(new HashMap<>((Map<String, Object>) metadataMap));
}
document.setScore(score);
return document;
}
Map<String, Object> buildSource(Document document) {
Map<String, Object> source = new HashMap<String, Object>();
source.put("id", document.getId());
source.put("content", document.getContent());
if (document.getTitle() != null) {
source.put("title", document.getTitle());
}
if (document.getMetadataMap() != null && !document.getMetadataMap().isEmpty()) {
source.put("metadataMap", new HashMap<String, Object>(document.getMetadataMap()));
Object knowledgeId = document.getMetadata(KeywordSearchMetadataKeys.KNOWLEDGE_ID);
if (knowledgeId != null) {
source.put(KeywordSearchMetadataKeys.KNOWLEDGE_ID, String.valueOf(knowledgeId));
}
}
return source;
}
SearchRequest buildSearchRequest(KeywordSearchRequest request) {
KeywordSearchRequest effectiveRequest = request == null ? new KeywordSearchRequest() : request;
return SearchRequest.of(s -> s
.index(esConfig.getIndexName())
.size(effectiveRequest.getCount())
.source(SourceConfig.of(sc -> sc.filter(f -> f.includes("id", "title", "content", "metadataMap"))))
.query(q -> q.bool(b -> {
b.must(m -> m.multiMatch(mm -> mm
.query(effectiveRequest.getKeyword())
.fields("title", "content")
));
if (effectiveRequest.getKnowledgeId() != null && !effectiveRequest.getKnowledgeId().trim().isEmpty()) {
b.filter(f -> f.term(t -> t
.field(KeywordSearchMetadataKeys.KNOWLEDGE_ID)
.value(v -> v.stringValue(effectiveRequest.getKnowledgeId().trim()))
));
}
return b;
}))
);
}
public boolean checkAvailable() {
RestClient restClient = null;
ElasticsearchTransport transport = null;
try {
restClient = buildRestClient();
transport = new RestClientTransport(restClient, new JacksonJsonpMapper());
ElasticsearchClient client = new ElasticsearchClient(transport);
return client.info() != null;
} catch (Exception e) {
LOG.error("Elasticsearch availability check failed", e);
return false;
} finally {
closeResources(transport, restClient);
}
}
}

View File

@@ -0,0 +1,54 @@
package com.easyagents.engine.es;
import co.elastic.clients.elasticsearch.core.SearchRequest;
import com.easyagents.core.document.Document;
import com.easyagents.search.engine.service.KeywordSearchMetadataKeys;
import com.easyagents.search.engine.service.KeywordSearchRequest;
import org.junit.Assert;
import org.junit.Test;
import java.util.Map;
public class ElasticSearcherQueryBuilderTest {
@Test
public void shouldBuildSearchRequestWithMultiMatchAndKnowledgeFilter() {
ElasticSearcher searcher = new ElasticSearcher(config());
KeywordSearchRequest request = KeywordSearchRequest.of("客服", 5);
request.setKnowledgeId("100");
SearchRequest searchRequest = searcher.buildSearchRequest(request);
Assert.assertEquals(5, searchRequest.size().intValue());
Assert.assertNotNull(searchRequest.query().bool());
Assert.assertEquals(1, searchRequest.query().bool().must().size());
Assert.assertNotNull(searchRequest.query().bool().must().get(0).multiMatch());
Assert.assertEquals(2, searchRequest.query().bool().must().get(0).multiMatch().fields().size());
Assert.assertEquals(1, searchRequest.query().bool().filter().size());
Assert.assertEquals("knowledgeId", searchRequest.query().bool().filter().get(0).term().field());
}
@Test
public void shouldExtractKnowledgeIdToTopLevelSource() {
ElasticSearcher searcher = new ElasticSearcher(config());
Document document = new Document();
document.setId("1");
document.setTitle("title");
document.setContent("content");
document.addMetadata(KeywordSearchMetadataKeys.KNOWLEDGE_ID, "100");
Map<String, Object> source = searcher.buildSource(document);
Assert.assertEquals("100", source.get(KeywordSearchMetadataKeys.KNOWLEDGE_ID));
Assert.assertTrue(source.get("metadataMap") instanceof Map);
}
private ESConfig config() {
ESConfig config = new ESConfig();
config.setHost("http://127.0.0.1:9200");
config.setUserName("elastic");
config.setPassword("elastic");
config.setIndexName("easyflow");
return config;
}
}

View File

@@ -51,5 +51,10 @@
<groupId>com.easyagents</groupId>
<artifactId>easy-agents-search-engine-service</artifactId>
</dependency>
<dependency>
<groupId>junit</groupId>
<artifactId>junit</artifactId>
<scope>test</scope>
</dependency>
</dependencies>
</project>

View File

@@ -17,6 +17,8 @@ package com.easyagents.search.engine.lucene;
import com.easyagents.core.document.Document;
import com.easyagents.search.engine.service.DocumentSearcher;
import com.easyagents.search.engine.service.KeywordSearchMetadataKeys;
import com.easyagents.search.engine.service.KeywordSearchRequest;
import org.apache.lucene.analysis.Analyzer;
import org.apache.lucene.document.Field;
import org.apache.lucene.document.StringField;
@@ -78,7 +80,7 @@ public class LuceneSearcher implements DocumentSearcher {
if (document.getTitle() != null) {
luceneDoc.add(new TextField("title", document.getTitle(), Field.Store.YES));
}
appendKnowledgeId(document, luceneDoc);
indexWriter.addDocument(luceneDoc);
indexWriter.commit();
@@ -127,7 +129,7 @@ public class LuceneSearcher implements DocumentSearcher {
if (document.getTitle() != null) {
luceneDoc.add(new TextField("title", document.getTitle(), Field.Store.YES));
}
appendKnowledgeId(document, luceneDoc);
indexWriter.updateDocument(term, luceneDoc);
indexWriter.commit();
return true;
@@ -140,18 +142,21 @@ public class LuceneSearcher implements DocumentSearcher {
}
@Override
public List<Document> searchDocuments(String keyword, int count) {
public List<Document> searchDocuments(KeywordSearchRequest request) {
List<Document> results = new ArrayList<>();
try (IndexReader reader = DirectoryReader.open(directory)) {
IndexSearcher searcher = new IndexSearcher(reader);
Query query = buildQuery(keyword);
TopDocs topDocs = searcher.search(query, count);
Query query = buildQuery(request);
TopDocs topDocs = searcher.search(query, request == null ? 10 : request.getCount());
for (ScoreDoc scoreDoc : topDocs.scoreDocs) {
org.apache.lucene.document.Document doc = searcher.doc(scoreDoc.doc);
Document resultDoc = new Document();
resultDoc.setId(doc.get("id"));
resultDoc.setContent(doc.get("content"));
resultDoc.setTitle(doc.get("title"));
if (doc.get(KeywordSearchMetadataKeys.KNOWLEDGE_ID) != null) {
resultDoc.addMetadata(KeywordSearchMetadataKeys.KNOWLEDGE_ID, doc.get(KeywordSearchMetadataKeys.KNOWLEDGE_ID));
}
resultDoc.setScore((double) scoreDoc.score);
@@ -164,9 +169,10 @@ public class LuceneSearcher implements DocumentSearcher {
return results;
}
private static Query buildQuery(String keyword) {
Query buildQuery(KeywordSearchRequest request) {
try {
Analyzer analyzer = createAnalyzer();
String keyword = request == null ? null : request.getKeyword();
QueryParser titleQueryParser = new QueryParser("title", analyzer);
Query titleQuery = titleQueryParser.parse(keyword);
@@ -179,6 +185,9 @@ public class LuceneSearcher implements DocumentSearcher {
BooleanQuery.Builder builder = new BooleanQuery.Builder();
builder.add(titleBooleanClause)
.add(contentBooleanClause);
if (request != null && request.getKnowledgeId() != null && !request.getKnowledgeId().trim().isEmpty()) {
builder.add(new TermQuery(new Term(KeywordSearchMetadataKeys.KNOWLEDGE_ID, request.getKnowledgeId().trim())), BooleanClause.Occur.MUST);
}
return builder.build();
} catch (ParseException e) {
LOG.error(e.toString(), e);
@@ -200,6 +209,16 @@ public class LuceneSearcher implements DocumentSearcher {
return new JcsegAnalyzer(ISegment.Type.NLP, config, DictionaryFactory.createSingletonDictionary(config));
}
private void appendKnowledgeId(Document document, org.apache.lucene.document.Document luceneDoc) {
if (document == null || document.getMetadataMap() == null) {
return;
}
Object knowledgeId = document.getMetadata(KeywordSearchMetadataKeys.KNOWLEDGE_ID);
if (knowledgeId != null) {
luceneDoc.add(new StringField(KeywordSearchMetadataKeys.KNOWLEDGE_ID, String.valueOf(knowledgeId), Field.Store.YES));
}
}
public void close(IndexWriter indexWriter) {
try {
if (indexWriter != null) {

View File

@@ -0,0 +1,45 @@
package com.easyagents.search.engine.lucene;
import com.easyagents.core.document.Document;
import com.easyagents.search.engine.service.KeywordSearchMetadataKeys;
import com.easyagents.search.engine.service.KeywordSearchRequest;
import org.junit.Assert;
import org.junit.Test;
import java.nio.file.Files;
import java.nio.file.Path;
import java.util.List;
public class LuceneSearcherTest {
@Test
public void shouldFilterByKnowledgeIdAndSearchTitleAndContent() throws Exception {
Path tempDir = Files.createTempDirectory("lucene-searcher-test");
LuceneConfig config = new LuceneConfig();
config.setIndexDirPath(tempDir.toString());
LuceneSearcher searcher = new LuceneSearcher(config);
Document first = new Document();
first.setId("1");
first.setTitle("客服标题");
first.setContent("这里没有关键字");
first.addMetadata(KeywordSearchMetadataKeys.KNOWLEDGE_ID, "100");
Document second = new Document();
second.setId("2");
second.setTitle("别的知识库");
second.setContent("客服内容");
second.addMetadata(KeywordSearchMetadataKeys.KNOWLEDGE_ID, "200");
Assert.assertTrue(searcher.addDocument(first));
Assert.assertTrue(searcher.addDocument(second));
KeywordSearchRequest request = KeywordSearchRequest.of("客服", 10);
request.setKnowledgeId("100");
List<Document> results = searcher.searchDocuments(request);
Assert.assertEquals(1, results.size());
Assert.assertEquals("1", String.valueOf(results.get(0).getId()));
Assert.assertEquals("100", String.valueOf(results.get(0).getMetadata(KeywordSearchMetadataKeys.KNOWLEDGE_ID)));
}
}

View File

@@ -28,8 +28,12 @@ public interface DocumentSearcher {
boolean updateDocument(Document document);
default List<Document> searchDocuments(String keyword) {
return searchDocuments(keyword, 10);
return searchDocuments(KeywordSearchRequest.of(keyword, 10));
}
List<Document> searchDocuments(String keyword, int count);
default List<Document> searchDocuments(String keyword, int count) {
return searchDocuments(KeywordSearchRequest.of(keyword, count));
}
List<Document> searchDocuments(KeywordSearchRequest request);
}

View File

@@ -0,0 +1,9 @@
package com.easyagents.search.engine.service;
public final class KeywordSearchMetadataKeys {
private KeywordSearchMetadataKeys() {
}
public static final String KNOWLEDGE_ID = "knowledgeId";
}

View File

@@ -0,0 +1,39 @@
package com.easyagents.search.engine.service;
public class KeywordSearchRequest {
private String keyword;
private int count = 10;
private String knowledgeId;
public static KeywordSearchRequest of(String keyword, int count) {
KeywordSearchRequest request = new KeywordSearchRequest();
request.setKeyword(keyword);
request.setCount(count);
return request;
}
public String getKeyword() {
return keyword;
}
public void setKeyword(String keyword) {
this.keyword = keyword;
}
public int getCount() {
return count;
}
public void setCount(int count) {
this.count = count <= 0 ? 10 : count;
}
public String getKnowledgeId() {
return knowledgeId;
}
public void setKnowledgeId(String knowledgeId) {
this.knowledgeId = knowledgeId;
}
}