Merge branch 'feature/luojunhui/20260615-add-article-text-summary-embedding' of Server/video-vector-server into master

core/src/main/java/com/tzld/videoVector/common/constant/VectorConstants.java

@@ -16,6 +16,12 @@ public interface VectorConstants {
     /** 视频标题配置编码 */
     String VIDEO_TITLE_CONFIG_CODE = "VIDEO_TITLE";
 
+    /** 文章标题配置编码 */
+    String ARTICLE_TITLE_CONFIG_CODE = "ARTICLE_TITLE";
+
+    /** 文章摘要配置编码 */
+    String ARTICLE_SUMMARY_CONFIG_CODE = "ARTICLE_SUMMARY";
+
     /** configCode 传 "ALL" 表示搜索所有启用的向量化配置 */
     String ALL_CONFIG_CODE = "ALL";
 

core/src/main/java/com/tzld/videoVector/dao/mapper/pgVector/ext/ArticleDeconstructResultMapperExt.java

@@ -21,4 +21,6 @@ public interface ArticleDeconstructResultMapperExt {
 
     List<ArticleDeconstructResult> selectResultsByArticleIds(@Param("source") String source,
                                                                @Param("articleIds") List<String> articleIds);
+
+    int upsertOdpsText(@Param("list") List<ArticleDeconstructResult> list);
 }

core/src/main/java/com/tzld/videoVector/job/ArticleTextVectorJob.java

@@ -0,0 +1,395 @@
+package com.tzld.videoVector.job;
+
+import com.alibaba.fastjson.JSONObject;
+import com.tzld.videoVector.common.constant.VectorConstants;
+import com.tzld.videoVector.dao.mapper.pgVector.DeconstructVectorConfigMapper;
+import com.tzld.videoVector.dao.mapper.pgVector.ext.ArticleDeconstructResultMapperExt;
+import com.tzld.videoVector.model.po.pgVector.ArticleDeconstructResult;
+import com.tzld.videoVector.model.po.pgVector.DeconstructVectorConfig;
+import com.tzld.videoVector.model.po.pgVector.DeconstructVectorConfigExample;
+import com.tzld.videoVector.service.ArticleVectorStoreService;
+import com.tzld.videoVector.service.EmbeddingService;
+import com.tzld.videoVector.util.Md5Util;
+import com.tzld.videoVector.util.OdpsUtil;
+import com.xxl.job.core.biz.model.ReturnT;
+import com.xxl.job.core.handler.annotation.XxlJob;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.beans.factory.annotation.Value;
+import org.springframework.stereotype.Component;
+import org.springframework.util.CollectionUtils;
+import org.springframework.util.StringUtils;
+
+import javax.annotation.Resource;
+import java.util.ArrayList;
+import java.util.HashSet;
+import java.util.LinkedHashMap;
+import java.util.List;
+import java.util.Map;
+import java.util.Set;
+
+/**
+ * 文章标题/摘要向量化定时任务（独立于 AIGC 解构流水线）
+ * <p>
+ * 数据源：ODPS 大数据表（通过 content_id 关联文章标题和摘要）
+ * 存储目标：article_vectors 表（configCode = ARTICLE_TITLE / ARTICLE_SUMMARY）
+ * <p>
+ * 设计参考 {@link VideoTitleVectorJob}：独立 Job → 外部数据源获取文本 → embedding → 写入向量表
+ */
+@Slf4j
+@Component
+public class ArticleTextVectorJob {
+
+    private static final String SOURCE_AIGC = "aigc_deconstruct";
+    private static final String SOURCE_ODPS_TEXT = "odps_text";
+
+    /** 每批 DB 写入数量 */
+    private static final int DB_BATCH_SIZE = 200;
+
+    @Resource
+    private DeconstructVectorConfigMapper vectorConfigMapper;
+
+    @Resource
+    private ArticleVectorStoreService articleVectorStoreService;
+
+    @Resource
+    private EmbeddingService embeddingService;
+
+    @Resource
+    private ArticleDeconstructResultMapperExt articleDeconstructResultMapperExt;
+
+    // ==================== ODPS 配置 ====================
+
+    @Value("${odps.article.text.table:loghubods.long_articles_text_hour}")
+    private String odpsTable;
+
+    @Value("${odps.article.text.id.column:content_id}")
+    private String idColumn;
+
+    @Value("${odps.article.text.title.column:article_title}")
+    private String titleColumn;
+
+    @Value("${odps.article.text.summary.column:kimi_summary}")
+    private String summaryColumn;
+
+    // ==================== 处理参数 ====================
+
+    /** embedding 调用间隔（毫秒），防止超过 API 频次限制 */
+    @Value("${article.text.vector.embedding.interval.ms:100}")
+    private long embeddingIntervalMs;
+
+    // ====================================================================
+    // XXL-Job 入口
+    // ====================================================================
+
+    @XxlJob("articleTextVectorJob")
+    public ReturnT<String> articleTextVectorJob(String param) {
+        log.info("开始执行文章标题/摘要向量化任务, param: {}", param);
+
+        String dt = parseParamString(param, "dt", null);
+
+        try {
+            // 1. 加载配置
+            DeconstructVectorConfig titleConfig = getConfigByCode(VectorConstants.ARTICLE_TITLE_CONFIG_CODE);
+            DeconstructVectorConfig summaryConfig = getConfigByCode(VectorConstants.ARTICLE_SUMMARY_CONFIG_CODE);
+            if (titleConfig == null && summaryConfig == null) {
+                log.error("未找到 ARTICLE_TITLE / ARTICLE_SUMMARY 向量化配置，请先插入 deconstruct_vector_config 记录");
+                return new ReturnT<>(ReturnT.FAIL_CODE, "未找到配置");
+            }
+            logConfig(titleConfig, "ARTICLE_TITLE");
+            logConfig(summaryConfig, "ARTICLE_SUMMARY");
+
+            // 2. 加载已入库的 articleId 集合（从 article_deconstruct_result 分页扫描）
+            Set<String> allArticleIds = loadAllArticleIds();
+            log.info("article_deconstruct_result 中共 {} 个 articleId", allArticleIds.size());
+            if (allArticleIds.isEmpty()) {
+                log.info("无文章数据，任务结束");
+                return ReturnT.SUCCESS;
+            }
+
+            // 3. 从 ODPS 流式读取标题/摘要，仅保留已入库文章
+            Map<String, String> titleMap = new LinkedHashMap<>();
+            Map<String, String> summaryMap = new LinkedHashMap<>();
+            long odpsTotal = loadTextFromOdps(dt, allArticleIds, titleMap, summaryMap);
+            log.info("ODPS 流式读取完成, 总行数={}, 匹配标题={}, 匹配摘要={}",
+                    odpsTotal, titleMap.size(), summaryMap.size());
+
+            // 3.5 持久化 ODPS 标题/摘要到 article_deconstruct_result（source=odps_text）
+            int persisted = persistOdpsTextToDb(titleMap, summaryMap);
+            log.info("ODPS 文本持久化完成, 写入 {} 条", persisted);
+
+            // 4. 向量化标题
+            if (titleConfig != null) {
+                processTextVector(titleConfig, titleMap, "标题");
+            }
+
+            // 5. 向量化摘要
+            if (summaryConfig != null) {
+                processTextVector(summaryConfig, summaryMap, "摘要");
+            }
+
+            log.info("文章标题/摘要向量化任务完成");
+            return ReturnT.SUCCESS;
+        } catch (Exception e) {
+            log.error("文章标题/摘要向量化任务失败: {}", e.getMessage(), e);
+            return new ReturnT<>(ReturnT.FAIL_CODE, "任务执行失败: " + e.getMessage());
+        }
+    }
+
+    // ====================================================================
+    // 文章 ID 加载
+    // ====================================================================
+
+    private Set<String> loadAllArticleIds() {
+        Set<String> ids = new HashSet<>();
+        int pageNum = 0;
+        while (true) {
+            int offset = pageNum * VectorConstants.PAGE_SIZE;
+            List<String> page = articleDeconstructResultMapperExt
+                    .selectArticleIdsBySourcePaged(SOURCE_AIGC, offset, VectorConstants.PAGE_SIZE);
+            if (CollectionUtils.isEmpty(page)) {
+                break;
+            }
+            ids.addAll(page);
+            if (page.size() < VectorConstants.PAGE_SIZE) {
+                break;
+            }
+            pageNum++;
+        }
+        return ids;
+    }
+
+    // ====================================================================
+    // ODPS 数据加载
+    // ====================================================================
+
+    private long loadTextFromOdps(String dt, Set<String> allArticleIds,
+                                   Map<String, String> titleMap, Map<String, String> summaryMap) {
+        String safeDt = dt != null ? dt.replace("'", "''") : null;
+        String safeIdCol = idColumn.replace("'", "''");
+        String safeTitleCol = titleColumn.replace("'", "''");
+        String safeSummaryCol = summaryColumn.replace("'", "''");
+        String safeTable = odpsTable.replace("'", "''");
+
+        StringBuilder sql = new StringBuilder();
+        sql.append("SELECT ").append(safeIdCol)
+           .append(", ").append(safeTitleCol)
+           .append(", ").append(safeSummaryCol)
+           .append(" FROM ").append(safeTable);
+        if (safeDt != null) {
+            sql.append(" WHERE dt = '").append(safeDt).append("'");
+        }
+
+        log.info("ODPS SQL: {}", sql);
+
+        long[] totalRows = {0};
+        long[] matchedRows = {0};
+
+        OdpsUtil.getOdpsDataStream(sql.toString(), record -> {
+            String contentId = record.getString(safeIdCol);
+            if (contentId == null || contentId.isEmpty()) {
+                totalRows[0]++;
+                return;
+            }
+            contentId = contentId.trim();
+
+            // 采样日志
+            if (totalRows[0] < 3) {
+                String t = record.getString(safeTitleCol);
+                String s = record.getString(safeSummaryCol);
+                log.info("[ODPS采样{}] contentId={}, title={}, summary={}",
+                        totalRows[0], contentId,
+                        t != null && t.length() > 100 ? t.substring(0, 100) + "..." : t,
+                        s != null && s.length() > 100 ? s.substring(0, 100) + "..." : s);
+            }
+
+            if (allArticleIds.contains(contentId)) {
+                String title = record.getString(safeTitleCol);
+                if (StringUtils.hasText(title)) {
+                    titleMap.put(contentId, title.trim());
+                }
+                String summary = record.getString(safeSummaryCol);
+                if (StringUtils.hasText(summary)) {
+                    summaryMap.put(contentId, summary.trim());
+                }
+                matchedRows[0]++;
+            }
+            totalRows[0]++;
+
+            if (totalRows[0] % 10000 == 0) {
+                log.info("[ODPS进度] {} 行, 匹配={}", totalRows[0], matchedRows[0]);
+            }
+        });
+
+        return totalRows[0];
+    }
+
+    // ====================================================================
+    // 持久化 ODPS 文本到 article_deconstruct_result
+    // ====================================================================
+
+    /**
+     * 将 ODPS 标题/摘要以 source=odps_text 写入 article_deconstruct_result。
+     * result 字段存储精简 JSON：{"title":"...", "summary":"..."}
+     */
+    private int persistOdpsTextToDb(Map<String, String> titleMap, Map<String, String> summaryMap) {
+        Set<String> allIds = new LinkedHashSet<>();
+        allIds.addAll(titleMap.keySet());
+        allIds.addAll(summaryMap.keySet());
+        if (allIds.isEmpty()) {
+            return 0;
+        }
+
+        List<ArticleDeconstructResult> batch = new ArrayList<>(DB_BATCH_SIZE);
+        int totalPersisted = 0;
+
+        for (String articleId : allIds) {
+            JSONObject json = new JSONObject();
+            String title = titleMap.get(articleId);
+            if (StringUtils.hasText(title)) {
+                json.put("title", title);
+            }
+            String summary = summaryMap.get(articleId);
+            if (StringUtils.hasText(summary)) {
+                json.put("summary", summary);
+            }
+            if (json.isEmpty()) {
+                continue;
+            }
+
+            ArticleDeconstructResult row = new ArticleDeconstructResult();
+            row.setArticleId(articleId);
+            row.setSource(SOURCE_ODPS_TEXT);
+            row.setResult(json.toJSONString());
+            batch.add(row);
+
+            if (batch.size() >= DB_BATCH_SIZE) {
+                totalPersisted += articleDeconstructResultMapperExt.upsertOdpsText(batch);
+                batch.clear();
+            }
+        }
+
+        if (!batch.isEmpty()) {
+            totalPersisted += articleDeconstructResultMapperExt.upsertOdpsText(batch);
+        }
+
+        return totalPersisted;
+    }
+
+    // ====================================================================
+    // 向量化处理
+    // ====================================================================
+
+    private void processTextVector(DeconstructVectorConfig config, Map<String, String> textMap, String label) {
+        String configCode = config.getConfigCode();
+        Set<String> existingIds = articleVectorStoreService.getAllArticleIds(configCode);
+        log.info("{}向量化开始, 待处理={}, 已有向量={}", label, textMap.size(), existingIds.size());
+
+        Integer maxLength = config.getMaxLength();
+        int success = 0;
+        int fail = 0;
+        int skip = 0;
+        int processed = 0;
+
+        for (Map.Entry<String, String> entry : textMap.entrySet()) {
+            String articleId = entry.getKey();
+            String text = entry.getValue();
+
+            if (existingIds.contains(articleId)) {
+                skip++;
+                continue;
+            }
+            if (!StringUtils.hasText(text)) {
+                skip++;
+                continue;
+            }
+
+            // 截断
+            if (maxLength != null && maxLength > 0 && text.length() > maxLength) {
+                text = text.substring(0, maxLength);
+            }
+
+            try {
+                List<Float> vector = getOrEmbed(text, config);
+                if (vector == null || vector.isEmpty()) {
+                    log.error("{}向量化失败, articleId={}, text={}", label, articleId,
+                            text.length() > 50 ? text.substring(0, 50) + "..." : text);
+                    fail++;
+                    continue;
+                }
+
+                if (!articleVectorStoreService.save(configCode, articleId, vector, text)) {
+                    log.error("{}向量存储失败, articleId={}", label, articleId);
+                    fail++;
+                    continue;
+                }
+                success++;
+            } catch (Exception e) {
+                log.error("{}向量化异常, articleId={}: {}", label, articleId, e.getMessage(), e);
+                fail++;
+            }
+
+            processed++;
+            if (processed % 100 == 0) {
+                log.info("{}向量化进度: 成功={}, 失败={}, 跳过={}", label, success, fail, skip);
+            }
+        }
+
+        log.info("{}向量化完成: 成功={}, 失败={}, 跳过={}", label, success, fail, skip);
+    }
+
+    // ====================================================================
+    // Embedding（含 text_hash 缓存 + 限流）
+    // ====================================================================
+
+    private List<Float> getOrEmbed(String text, DeconstructVectorConfig config) {
+        String configCode = config.getConfigCode();
+        String textHash = Md5Util.encoderByMd5(text);
+        if (StringUtils.hasText(textHash)) {
+            List<Float> cached = articleVectorStoreService.getVectorByTextHash(textHash, configCode);
+            if (cached != null && !cached.isEmpty()) {
+                log.debug("命中 text_hash 缓存, hash={}, configCode={}", textHash, configCode);
+                return cached;
+            }
+        }
+        // 限流
+        try {
+            Thread.sleep(embeddingIntervalMs);
+        } catch (InterruptedException e) {
+            Thread.currentThread().interrupt();
+        }
+        return embeddingService.embed(text, config);
+    }
+
+    // ====================================================================
+    // 工具方法
+    // ====================================================================
+
+    private DeconstructVectorConfig getConfigByCode(String configCode) {
+        DeconstructVectorConfigExample example = new DeconstructVectorConfigExample();
+        example.createCriteria()
+                .andEnabledEqualTo((short) 1)
+                .andConfigCodeEqualTo(configCode);
+        List<DeconstructVectorConfig> configs = vectorConfigMapper.selectByExample(example);
+        return CollectionUtils.isEmpty(configs) ? null : configs.get(0);
+    }
+
+    private void logConfig(DeconstructVectorConfig config, String label) {
+        if (config == null) {
+            log.info("{} 配置未找到，跳过", label);
+            return;
+        }
+        log.info("{} 配置: model={}, dimension={}, maxLength={}",
+                label, config.getEmbeddingModel(), config.getDimension(), config.getMaxLength());
+    }
+
+    private static String parseParamString(String param, String key, String defaultValue) {
+        if (param == null || param.isEmpty()) return defaultValue;
+        for (String part : param.split(",")) {
+            String[] kv = part.trim().split("=", 2);
+            if (kv.length == 2 && kv[0].trim().equals(key)) {
+                return kv[1].trim();
+            }
+        }
+        return defaultValue;
+    }
+}

core/src/main/java/com/tzld/videoVector/model/vo/recall/ArticleBasicVO.java

@@ -14,7 +14,7 @@ public class ArticleBasicVO {
     /** 文章标题 */
     private String title;
 
-    /** 摘要（body_text 前 120 字符） */
+    /** 摘要（优先 ODPS 真实摘要，回退 body_text 前 200 字符） */
     private String summary;
 
     /** 封面图 CDN 地址 */

core/src/main/java/com/tzld/videoVector/service/recall/impl/VectorRecallTestServiceImpl.java

@@ -136,6 +136,7 @@ public class VectorRecallTestServiceImpl implements VectorRecallTestService {
     private int metricsDays;
 
     private static final String SOURCE_AIGC = "aigc_deconstruct";
+    private static final String SOURCE_ODPS_TEXT = "odps_text";
 
     /** source_type → 中文来源标签 */
     private static final short SOURCE_TYPE_EXTERNAL = 1;
@@ -356,6 +357,9 @@ public class VectorRecallTestServiceImpl implements VectorRecallTestService {
         log.info("batchByText: queryText={}, configCodes={}, recallK={}, displayK={}, enrichK={}",
                 queryText, configCodes, recallK, displayK, enrichK);
 
+        // 预加载 config → contentType 映射，避免按无关模态执行 ANN
+        Map<String, Short> ctByCode = loadContentTypeMap(configCodes);
+
         // ============================================================
         // 阶段一：平行 ANN（不做 enrich），收集原始匹配
         // ============================================================
@@ -370,7 +374,12 @@ public class VectorRecallTestServiceImpl implements VectorRecallTestService {
 
         for (String configCode : configCodes) {
             final String cc = configCode;
-            if (needVideo) {
+            Short ct = ctByCode.get(cc);
+            boolean doVideo = needVideo && appliesToModality(ct, (short) 3);
+            boolean doMaterial = needMaterial && appliesToModality(ct, (short) 2);
+            boolean doArticle = needArticle && appliesToModality(ct, (short) 1);
+
+            if (doVideo) {
                 futures.add(CompletableFuture.runAsync(() -> {
                     try {
                         MatchTopNVideoParam videoParam = new MatchTopNVideoParam();
@@ -387,7 +396,7 @@ public class VectorRecallTestServiceImpl implements VectorRecallTestService {
             }, RECALL_EXECUTOR));
             }
 
-            if (needMaterial) {
+            if (doMaterial) {
                 futures.add(CompletableFuture.runAsync(() -> {
                     try {
                         List<MaterialMatch> matches = rawMaterialANN(queryText, cc, recallK,
@@ -401,7 +410,7 @@ public class VectorRecallTestServiceImpl implements VectorRecallTestService {
                 }, RECALL_EXECUTOR));
             }
 
-            if (needArticle) {
+            if (doArticle) {
                 futures.add(CompletableFuture.runAsync(() -> {
                     try {
                         List<ArticleMatch> matches = rawArticleANN(queryText, cc, recallK);
@@ -656,6 +665,36 @@ public class VectorRecallTestServiceImpl implements VectorRecallTestService {
         }
     }
 
+    /**
+     * 加载 configCode → contentType 映射，用于 batchByText 按模态过滤 ANN。
+     * contentType: 1=长文, 2=图文, 3=视频, null=全部
+     */
+    private Map<String, Short> loadContentTypeMap(List<String> configCodes) {
+        if (CollectionUtils.isEmpty(configCodes)) {
+            return Collections.emptyMap();
+        }
+        try {
+            DeconstructVectorConfigExample example = new DeconstructVectorConfigExample();
+            example.createCriteria().andConfigCodeIn(configCodes);
+            List<DeconstructVectorConfig> configs = deconstructVectorConfigMapper.selectByExample(example);
+            Map<String, Short> map = new HashMap<>();
+            for (DeconstructVectorConfig c : configs) {
+                if (StringUtils.hasText(c.getConfigCode())) {
+                    map.put(c.getConfigCode(), c.getContentType());
+                }
+            }
+            return map;
+        } catch (Exception e) {
+            log.warn("加载 contentType 映射失败，将退化到全模态 ANN: {}", e.getMessage());
+            return Collections.emptyMap();
+        }
+    }
+
+    /** config 的 contentType 为 null（全部）或等于 targetCt 时，该模态需要执行 ANN。 */
+    private static boolean appliesToModality(Short contentType, short targetCt) {
+        return contentType == null || contentType == targetCt;
+    }
+
     private List<VideoMatchResult> limitVideoMatchesByScore(List<VideoMatchResult> matches, int topN) {
         if (CollectionUtils.isEmpty(matches) || topN <= 0 || matches.size() <= topN) {
             return matches == null ? Collections.emptyList() : matches;
@@ -1199,14 +1238,71 @@ public class VectorRecallTestServiceImpl implements VectorRecallTestService {
         try {
             List<ArticleDeconstructResult> rows = articleDeconstructResultMapperExt
                     .selectResultsByArticleIds(SOURCE_AIGC, articleIds);
-            if (CollectionUtils.isEmpty(rows)) {
-                return result;
+            if (!CollectionUtils.isEmpty(rows)) {
+                for (ArticleDeconstructResult row : rows) {
+                    if (row == null || !StringUtils.hasText(row.getArticleId())) {
+                        continue;
+                    }
+                    result.putIfAbsent(row.getArticleId(), row);
+                }
             }
-            for (ArticleDeconstructResult row : rows) {
-                if (row == null || !StringUtils.hasText(row.getArticleId())) {
-                    continue;
+
+            // 补充 ODPS 标题/摘要到 AIGC 解构 JSON 中
+            List<ArticleDeconstructResult> odpsRows = articleDeconstructResultMapperExt
+                    .selectResultsByArticleIds(SOURCE_ODPS_TEXT, articleIds);
+            if (!CollectionUtils.isEmpty(odpsRows)) {
+                for (ArticleDeconstructResult odpsRow : odpsRows) {
+                    if (odpsRow == null || !StringUtils.hasText(odpsRow.getArticleId())) {
+                        continue;
+                    }
+                    JSONObject odpsJson = parseArticleResultJson(odpsRow);
+                    if (odpsJson == null) continue;
+                    String odpsTitle = odpsJson.getString("title");
+                    String odpsSummary = odpsJson.getString("summary");
+
+                    ArticleDeconstructResult aigcRow = result.get(odpsRow.getArticleId());
+                    if (aigcRow != null && StringUtils.hasText(aigcRow.getResult())) {
+                        // 合并：将 ODPS 标题/摘要补充到 AIGC JSON 的 target_post 中
+                        try {
+                            JSONObject aigcJson = JSON.parseObject(aigcRow.getResult());
+                            JSONObject targetPost = aigcJson.getJSONObject("target_post");
+                            if (targetPost == null) {
+                                targetPost = new JSONObject();
+                                aigcJson.put("target_post", targetPost);
+                            }
+                            if (StringUtils.hasText(odpsTitle)
+                                    && !StringUtils.hasText(targetPost.getString("title"))) {
+                                targetPost.put("title", odpsTitle);
+                            }
+                            if (StringUtils.hasText(odpsSummary)
+                                    && !StringUtils.hasText(targetPost.getString("summary"))) {
+                                targetPost.put("summary", odpsSummary);
+                            }
+                            aigcRow.setResult(aigcJson.toJSONString());
+                        } catch (Exception e) {
+                            log.error("合并 ODPS 文本到 AIGC JSON 失败, articleId={}: {}",
+                                    odpsRow.getArticleId(), e.getMessage());
+                        }
+                    } else if (aigcRow == null) {
+                        // 仅 ODPS 数据、无 AIGC 解构结果：构造最小 JSON 供 enrich 使用
+                        JSONObject minimal = new JSONObject();
+                        JSONObject targetPost = new JSONObject();
+                        if (StringUtils.hasText(odpsTitle)) {
+                            targetPost.put("title", odpsTitle);
+                        }
+                        if (StringUtils.hasText(odpsSummary)) {
+                            targetPost.put("summary", odpsSummary);
+                        }
+                        if (!targetPost.isEmpty()) {
+                            minimal.put("target_post", targetPost);
+                            ArticleDeconstructResult newRow = new ArticleDeconstructResult();
+                            newRow.setArticleId(odpsRow.getArticleId());
+                            newRow.setSource(SOURCE_AIGC);
+                            newRow.setResult(minimal.toJSONString());
+                            result.put(odpsRow.getArticleId(), newRow);
+                        }
+                    }
                 }
-                result.putIfAbsent(row.getArticleId(), row);
             }
         } catch (Exception e) {
             log.error("批量加载 article_deconstruct_result 失败: {}", e.getMessage(), e);
@@ -1263,9 +1359,14 @@ public class VectorRecallTestServiceImpl implements VectorRecallTestService {
 
         ArticleBasicMeta meta = new ArticleBasicMeta();
         meta.title = targetPost.getString("title");
-        String bodyText = targetPost.getString("body_text");
-        if (StringUtils.hasText(bodyText)) {
-            meta.summary = bodyText.length() > 200 ? bodyText.substring(0, 200) : bodyText;
+        String summary = targetPost.getString("summary");
+        if (StringUtils.hasText(summary)) {
+            meta.summary = summary;
+        } else {
+            String bodyText = targetPost.getString("body_text");
+            if (StringUtils.hasText(bodyText)) {
+                meta.summary = bodyText.length() > 200 ? bodyText.substring(0, 200) : bodyText;
+            }
         }
         meta.articleId = targetPost.getString("channel_content_id");
 

core/src/main/resources/mapper/pgVector/ext/ArticleDeconstructResultMapperExt.xml

@@ -48,4 +48,15 @@
         </foreach>
     </select>
 
+    <insert id="upsertOdpsText">
+        INSERT INTO article_deconstruct_result (article_id, source, result)
+        VALUES
+        <foreach collection="list" item="item" separator=",">
+            (#{item.articleId}, #{item.source}, #{item.result})
+        </foreach>
+        ON CONFLICT (article_id, source) DO UPDATE SET
+            result = EXCLUDED.result,
+            update_time = NOW()
+    </insert>
+
 </mapper>