本讲介绍原材料中的两种 Chain：ConversationalChain 和 ConversationalRetrievalChain，对照它们如何组合模型、对话记忆与检索。类名和 API 对应原文版本，使用时需核对实际依赖。
原文配图：链式调用与会话检索示意 1。
ConversationalChain示例
@GetMapping(value = "/chat/chain", produces = MediaType.APPLICATION_JSON_VALUE)
public ResponseEntity<String> chatChain(@RequestParam String query) {
try {
String responseText = ConversationalChain.builder()
.chatModel(ollamaChatModel)
.chatMemory(MessageWindowChatMemory.withMaxMessages(10))
.build()
.execute(query);
return ResponseEntity.ok(responseText);
} catch (Exception e) {
log.error("chatChain", e);
return ResponseEntity.ok("{\"error\":\"chatChain error: " + e.getMessage() + "\"}");
}
}
这段链式调用把模型配置、对话记忆和查询执行组织在一起。
ConversationalRetrievalChain示例
/**
* 示例用测试文本data.txt向量化存储（按行分隔）
* @return
*/
EmbeddingStore<TextSegment> getEmbeddingStore() {
Resource resource = resourceLoader.getResource("classpath:data.txt");
File file = null;
try {
file = resource.getFile();
} catch (IOException e) {
throw new RuntimeException(e);
}
String path = file.getAbsolutePath();
EmbeddingStore<TextSegment> embeddingStore = new InMemoryEmbeddingStore<>();
Document document = FileSystemDocumentLoader.loadDocument(path);
DocumentByLineSplitter splitter = new DocumentByLineSplitter(100, 0);
List<TextSegment> split = splitter.split(document);
for (TextSegment textSegment : split) {
Embedding embedding = ollamaEmbeddingModel.embed(textSegment).content();
embeddingStore.add(embedding, textSegment);
}
return embeddingStore;
}
@GetMapping(value = "/rag/chain", produces = MediaType.APPLICATION_JSON_VALUE)
public ResponseEntity<String> ragChain(@RequestParam String query) {
try {
String answer = ConversationalRetrievalChain.builder()
.chatModel(ollamaChatModel)
.contentRetriever(EmbeddingStoreContentRetriever
.builder()
.embeddingModel(ollamaEmbeddingModel)
.embeddingStore(getEmbeddingStore())
.maxResults(1)
.build())
.chatMemory(MessageWindowChatMemory.withMaxMessages(10))
.build()
.execute(query);
return ResponseEntity.ok(answer);
} catch (Exception e) {
log.error("argChain", e);
return ResponseEntity.ok("{\"error\":\"argChain error: " + e.getMessage() + "\"}");
}
}
与ConversationalChain相比，ConversationalRetrievalChain新增了RAG的链式处理。
文本示例完整代码：https://github.com/yjmyzz/langchain4j-study/tree/day08