DevYangJC · tra371 · Jul 27, 2026
diff --git a/Argus-backend/src/main/java/com/argus/rag/engine/youcom/YoucomSearchService.java b/Argus-backend/src/main/java/com/argus/rag/engine/youcom/YoucomSearchService.java
@@ -0,0 +1,188 @@
+package com.argus.rag.engine.youcom;
+
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.beans.factory.annotation.Value;
+import org.springframework.stereotype.Service;
+import org.springframework.util.StringUtils;
+import org.springframework.ai.document.Document;
+
+import java.net.URI;
+import java.net.http.HttpClient;
+import java.net.http.HttpRequest;
+import java.net.http.HttpResponse;
+import java.time.Duration;
+import java.util.ArrayList;
+import java.util.LinkedHashMap;
+import java.util.List;
+import java.util.Map;
+
+/**
+ * You.com 网络搜索服务。
+ * <p>
+ * 通过 YDC Index API ({@code https://ydc-index.io/v1/search}) 执行网络搜索，
+ * 将搜索结果转换为 Spring AI {@link Document} 对象，
+ * 追加到检索证据束中作为第三条检索通道。
+ * </p>
+ *
+ * <p>通信方式：通过 JDK 原生 {@link HttpClient} 直接调用 YDC Index REST API。</p>
+ *
+ * @author Argus-RAG Team
+ */
+@Service
+@Slf4j
+public class YoucomSearchService {
+
+    private static final Duration REQUEST_TIMEOUT = Duration.ofSeconds(10);
+    private static final String SEARCH_URL = "https://ydc-index.io/v1/search";
+
+    private final HttpClient httpClient;
+    private final String apiKey;
+    private final boolean enabled;
+    private final int topK;
+    private final com.fasterxml.jackson.databind.ObjectMapper objectMapper;
+
+    /**
+     * 构造 You.com 搜索服务。
+     *
+     * @param apiKey You.com API 密钥，从环境变量或配置读取
+     * @param enabled 是否启用 You.com 搜索通道
+     * @param topK 每次搜索返回的最大结果数
+     * @param objectMapper Jackson ObjectMapper，用于 JSON 解析
+     */
+    @Autowired
+    public YoucomSearchService(
+            @Value("${youcom.search.api-key:}") String apiKey,
+            @Value("${youcom.search.enabled:false}") boolean enabled,
+            @Value("${youcom.search.top-k:10}") int topK,
+            com.fasterxml.jackson.databind.ObjectMapper objectMapper
+    ) {
+        this.httpClient = HttpClient.newBuilder().connectTimeout(REQUEST_TIMEOUT).build();
+        this.apiKey = apiKey;
+        this.enabled = enabled;
+        this.topK = Math.max(1, topK);
+        this.objectMapper = objectMapper;
+    }
+
+    /**
+     * 执行 You.com 网络搜索。
+     *
+     * @param query 搜索关键词 / 问题文本
+     * @return 搜索结果列表，每个结果为 Spring AI {@link Document}，content 取自 snippet，url 写入 metadata
+     */
+    public List<Document> search(String query) {
+        if (!enabled || !StringUtils.hasText(apiKey)) {
+            log.debug("You.com search disabled or API key not configured, skipping");
+            return List.of();
+        }
+
+        if (!StringUtils.hasText(query)) {
+            return List.of();
+        }
+
+        long startNano = System.nanoTime();
+        try {
+            String requestBody = """
+                    {
+                        "query": %s,
+                        "count": %d
+                    }
+                    """.formatted(toJsonString(query), topK);
+
+            HttpRequest request = HttpRequest.newBuilder()
+                    .uri(URI.create(SEARCH_URL))
+                    .header("Content-Type", "application/json")
+                    .header("Accept", "application/json")
+                    .header("X-API-Key", apiKey)
+                    .POST(HttpRequest.BodyPublishers.ofString(requestBody))
+                    .timeout(REQUEST_TIMEOUT)
+                    .build();
+
+            HttpResponse<String> response = httpClient.send(request, HttpResponse.BodyHandlers.ofString());
+            int status = response.statusCode();
+            if (status == 429) {
+                log.warn("You.com search rate limit exceeded (429)");
+                return List.of();
+            }
+            if (status == 401) {
+                log.warn("You.com API key invalid or expired (401)");
+                return List.of();
+            }
+            if (status != 200) {
+                log.warn("You.com search failed with status {}: {}", status, response.body());
+                return List.of();
+            }
+
+            List<Document> documents = parseResponse(response.body());
+            long elapsedMs = (System.nanoTime() - startNano) / 1_000_000;
+            log.info("You.com search completed: query={}, results={}, elapsedMs={}",
+                    query, documents.size(), elapsedMs);
+            return documents;
+
+        } catch (InterruptedException e) {
+            Thread.currentThread().interrupt();
+            log.error("You.com search interrupted: {}", e.getMessage());
+            return List.of();
+        } catch (Exception e) {
+            long elapsedMs = (System.nanoTime() - startNano) / 1_000_000;
+            log.error("You.com search error: {} (elapsedMs={})", e.getMessage(), elapsedMs);
+            return List.of();
+        }
+    }
+
+    private List<Document> parseResponse(String responseBody) {
+        List<Document> documents = new ArrayList<>();
+        try {
+            var node = objectMapper.readTree(responseBody);
+            var results = node.path("results");
+            if (!results.isArray()) {
+                return documents;
+            }
+
+            int index = 1;
+            for (var item : results) {
+                String title = item.path("title").asText("");
+                String url = item.path("url").asText("");
+                String content = extractSnippet(item);
+                if (content.isEmpty()) {
+                    content = item.path("description").asText("");
+                }
+
+                Map<String, Object> metadata = new LinkedHashMap<>();
+                metadata.put("evidenceId", "WEB" + index);
+                metadata.put("retrievalSource", "YOUCOM");
+                metadata.put("score", 1.0 / index);
+                metadata.put("url", url);
+                metadata.put("title", title);
+                metadata.put("coverageMode", "WEB_SEARCH");
+
+                String evidenceText = "来源：" + title + "\nURL：" + url + "\n" + content;
+                documents.add(Document.builder()
+                        .id("WEB" + index)
+                        .text(evidenceText)
+                        .metadata(metadata)
+                        .build());
+                index++;
+            }
+        } catch (Exception e) {
+            log.error("Failed to parse You.com response: {}", e.getMessage());
+        }
+        return documents;
+    }
+
+    private String extractSnippet(com.fasterxml.jackson.databind.JsonNode item) {
+        var snippets = item.path("snippets");
+        if (snippets.isArray() && !snippets.isEmpty()) {
+            return snippets.get(0).asText("");
+        }
+        return "";
+    }
+
+    private String toJsonString(String text) {
+        try {
+            return objectMapper.writeValueAsString(text);
+        } catch (Exception e) {
+            return "\"" + text.replace("\\", "\\\\").replace("\"", "\\\"") + "\"";
+        }
+    }
+}
diff --git a/Argus-backend/src/main/java/com/argus/rag/qa/rag/HybridChunkRetrievalService.java b/Argus-backend/src/main/java/com/argus/rag/qa/rag/HybridChunkRetrievalService.java
@@ -12,6 +12,7 @@
 import com.argus.rag.qa.service.QueryPlanningService;
 import com.argus.rag.engine.elasticsearch.ElasticsearchChunkIndexService;
 import com.argus.rag.engine.pgvector.PgVectorRetrievalAdapter;
+import com.argus.rag.engine.youcom.YoucomSearchService;
 import lombok.extern.slf4j.Slf4j;
 import org.slf4j.LoggerFactory;
 import org.springframework.ai.document.Document;
@@ -24,17 +25,19 @@
 /**
  * 混合文档切片检索服务。
  * <p>
- * 核心检索引擎，融合向量语义检索和关键词检索两个通道，
- * 通过 RRF（Reciprocal Rank Fusion）算法融合排序，
+ * 核心检索引擎，融合向量语义检索、关键词检索和 You.com 网络搜索三个通道，
+ * 向量+关键词通过 RRF（Reciprocal Rank Fusion）算法融合排序，
+ * You.com 结果追加到证据列表作为第三条通道，
  * 并支持邻居窗口扩展和证据充分度评估。
  * </p>
  * <h3>检索流程</h3>
  * <ol>
  * <li>查询规划：由 {@link QueryPlanningService} 分析问题并生成检索语句</li>
- * <li>双通道检索：向量检索 + 关键词检索</li>
- * <li>RRF 融合排序：合并两通道结果并按 RRF 评分排序</li>
+ * <li>三通道并行检索：向量检索 + 关键词检索 + You.com 网络搜索</li>
+ * <li>RRF 融合排序：合并向量+关键词通道结果并按 RRF 评分排序</li>
  * <li>聚类分组：将连续的切片聚合为类簇</li>
  * <li>邻居窗口扩展：扩展上下文窗口以提供更完整的证据</li>
+ * <li>You.com 结果追加：将 Web 搜索结果追加到证据列表</li>
  * <li>证据充分度评估：根据检索结果评估证据质量</li>
  * </ol>
  */
@@ -58,6 +61,8 @@ public class HybridChunkRetrievalService {
     private final PgVectorRetrievalAdapter vectorRetrievalAdapter;
     /** Elasticsearch 关键词检索服务 */
     private final ElasticsearchChunkIndexService elasticsearchChunkIndexService;
+    /** You.com 网络搜索服务 */
+    private final YoucomSearchService youcomSearchService;
     /** 文档切片数据访问层 */
     private final DocumentChunkMapper documentChunkMapper;
     /** 查询规划服务 */
@@ -74,10 +79,11 @@ public class HybridChunkRetrievalService {
     public HybridChunkRetrievalService(
             PgVectorRetrievalAdapter vectorRetrievalAdapter,
             ElasticsearchChunkIndexService elasticsearchChunkIndexService,
+            YoucomSearchService youcomSearchService,
             DocumentChunkMapper documentChunkMapper,
             QueryPlanningService queryPlanningService,
             DocumentMapper documentMapper) {
-        this(vectorRetrievalAdapter, elasticsearchChunkIndexService, documentChunkMapper, queryPlanningService, documentMapper, DEFAULT_NEIGHBOR_WINDOW);
+        this(vectorRetrievalAdapter, elasticsearchChunkIndexService, youcomSearchService, documentChunkMapper, queryPlanningService, documentMapper, DEFAULT_NEIGHBOR_WINDOW);
     }
 
     /**
@@ -86,12 +92,14 @@ public HybridChunkRetrievalService(
     public HybridChunkRetrievalService(
             PgVectorRetrievalAdapter vectorRetrievalAdapter,
             ElasticsearchChunkIndexService elasticsearchChunkIndexService,
+            YoucomSearchService youcomSearchService,
             DocumentChunkMapper documentChunkMapper,
             QueryPlanningService queryPlanningService,
             DocumentMapper documentMapper,
             int neighborWindow) {
         this.vectorRetrievalAdapter = vectorRetrievalAdapter;
         this.elasticsearchChunkIndexService = elasticsearchChunkIndexService;
+        this.youcomSearchService = youcomSearchService;
         this.documentChunkMapper = documentChunkMapper;
         this.queryPlanningService = queryPlanningService;
         this.documentMapper = documentMapper;
@@ -101,7 +109,7 @@ public HybridChunkRetrievalService(
     /**
      * 执行混合检索，返回包含证据文档和证据等级的完整检索结果。
      * <p>
-     * 流程：查询规划 → 双通道检索 → RRF 融合排序 → 聚类分组 → 窗口扩展 → 证据评估。
+     * 流程：查询规划 → 三通道并行检索 → RRF 融合排序 → 聚类分组 → 窗口扩展 → You.com 结果追加 → 证据评估。
      * </p>
      *
      * @param groupId  群组 ID，限定检索范围
@@ -182,6 +190,26 @@ public RetrievedEvidenceBundle retrieve(Long groupId, String question, int topK)
             log.info("混合检索证据组装为空: groupId={}, elapsedMs={}", validGroupId, elapsedMs);
             return RetrievedEvidenceBundle.empty();
         }
+
+        // You.com 网络搜索作为第三条通道，并行执行后直接追加到 documents 列表
+        for (String plannedQuery : queryPlan.queries()) {
+            List<Document> webResults = youcomSearchService.search(plannedQuery);
+            for (Document webDoc : webResults) {
+                // 更新 evidenceId 以保持连续编号
+                String newId = "E" + evidenceIndex;
+                Map<String, Object> metadata = new LinkedHashMap<>(webDoc.getMetadata());
+                metadata.put("evidenceId", newId);
+                documents.add(Document.builder()
+                        .id(newId)
+                        .text(webDoc.getText())
+                        .metadata(metadata)
+                        .build());
+                evidenceIndex++;
+            }
+        }
+        log.info("You.com 搜索结果追加完成: groupId={}, webResults={}",
+                validGroupId, evidenceIndex - rankedClusters.size() - 1);
+
         EvidenceLevel evidenceLevel = evaluateEvidenceLevel(documents);
         long elapsedMs = (System.nanoTime() - startNano) / 1_000_000;
         log.info("混合检索完成: groupId={}, evidenceCount={}, evidenceLevel={}, elapsedMs={}",

diff --git a/Argus-backend/src/main/resources/application-dev.yml b/Argus-backend/src/main/resources/application-dev.yml
@@ -102,3 +102,9 @@ ingestion:
   vector:
     # 控制业务侧每次调用 VectorStore.add 的 chunk 数量，降低单次 Ollama embedding 请求压力。
     add-batch-size: 9
+
+youcom:
+  search:
+    enabled: false
+    api-key: ${YOUCOM_API_KEY:}
+    top-k: 10