tt-help-cli-ycl 1.3.92 → 1.3.93

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tt-help-cli-ycl",
-  "version": "1.3.92",
+  "version": "1.3.93",
   "description": "TikTok user & video data scraper - extract ttSeller, verified, locationCreated from HTML source",
   "type": "module",
   "bin": {

package/src/cli/tag.js

@@ -619,9 +619,6 @@ export async function handleScoreAll(parsed) {
         });
         videos = enriched.videos;
 
-        // 更新 meta 中当前正在处理的标签
-        clientMeta.tag = tag;
-
         // 过滤 + 算分 (共用函数)
         const { matchedAuthorSet } = applyFilterAndScore(
           videos,

package/src/watch/data-store.js

@@ -2871,6 +2871,8 @@ Standards: 90-100=clear match, 70-89=likely, 50-69=possible, 20-49=low, 0-19=unl
     // 如果启用 LLM 打分，先采样一批进行评分（累积模式：按猜测国家分组，使用偏移量记忆避免重复采样）
     if (useLlm && normalizedLocations && normalizedLocations.length > 0) {
       const llmMinReturn = options.llmMinReturn ?? 60; // 最少返回合格数
+      const llmMinTagReturn = options.llmMinTagReturn ?? 30; // tag 最少合格数
+      const llmMinNonTagReturn = options.llmMinNonTagReturn ?? 30; // 非 tag 最少合格数
       const maxBatches = options.llmMaxBatches ?? 10; // 最多采样轮次，防止无限循环
 
       // 打印当前偏移量状态
@@ -2878,7 +2880,7 @@ Standards: 90-100=clear match, 70-89=likely, 50-69=possible, 20-49=low, 0-19=unl
         .map(([k, v]) => `${k}:${v}`)
         .join(", ");
       console.error(
-        `[data-store] LLM 打分开始: 符合条件 ${count} 条，每批 ${llmSampleSize} 条，最低分 ${llmMinScore}，最少返回 ${llmMinReturn} 条`,
+        `[data-store] LLM 打分开始: 符合条件 ${count} 条，每批 ${llmSampleSize} 条，最低分 ${llmMinScore}，tag 最少 ${llmMinTagReturn}，非 tag 最少 ${llmMinNonTagReturn}`,
       );
       if (offsetSummary) {
         console.error(`[data-store] 偏移量记忆: ${offsetSummary}`);
@@ -2886,7 +2888,8 @@ Standards: 90-100=clear match, 70-89=likely, 50-69=possible, 20-49=low, 0-19=unl
 
       // 返回 Promise，调用方需要 await
       return (async () => {
-        const allQualified = [];
+        const allTagQualified = []; // tag 合格列表（直接合格）
+        const allNonTagQualified = []; // 非 tag 合格列表（LLM 打分合格）
         const allScores = [];
 
         // 按猜测国家分组处理，每个国家使用独立的偏移量
@@ -2949,7 +2952,7 @@ Standards: 90-100=clear match, 70-89=likely, 50-69=possible, 20-49=low, 0-19=unl
 
             // tag 来源直接加入合格列表
             if (tagSamples.length > 0) {
-              allQualified.push(...tagSamples.map((s) => s.unique_id));
+              allTagQualified.push(...tagSamples.map((s) => s.unique_id));
               console.error(
                 `[data-store] ${location}: 本批 ${tagSamples.length} 条 tag 来源任务跳过 LLM 打分直接合格`,
               );
@@ -2964,47 +2967,51 @@ Standards: 90-100=clear match, 70-89=likely, 50-69=possible, 20-49=low, 0-19=unl
                 DEFAULT_TARGET_LOCATIONS,
               );
               batchQualified = scores.filter((s) => s.score >= llmMinScore);
+              allNonTagQualified.push(...batchQualified.map((s) => s.uniqueId));
             }
 
             allScores.push(...scores);
-            allQualified.push(...batchQualified.map((s) => s.uniqueId));
 
             totalBatches++;
+            const totalQualified = allTagQualified.length + allNonTagQualified.length;
             console.error(
-              `[data-store] ${location} 第 ${batch + 1} 批: 采样 ${samples.length} 条，本批合格 ${batchQualified.length} 条，累计合格 ${allQualified.length} 条`,
+              `[data-store] ${location} 第 ${batch + 1} 批: 采样 ${samples.length} 条，tag 合格 ${allTagQualified.length}，非 tag 合格 ${allNonTagQualified.length}，累计 ${totalQualified} 条`,
             );
 
             // 更新偏移量记忆
             offset += samples.length;
             llmSampleOffsets.set(location, offset);
 
-            // 合格数已达到最小返回阈值，停止采样
-            if (allQualified.length >= llmMinReturn) break;
+            // 检查是否两个类型都达到阈值，都达到才停止
+            const tagReached = allTagQualified.length >= llmMinTagReturn;
+            const nonTagReached = allNonTagQualified.length >= llmMinNonTagReturn;
+            if (tagReached && nonTagReached) {
+              console.error(
+                `[data-store] 两类任务均已达标 (tag: ${allTagQualified.length}/${llmMinTagReturn}, 非 tag: ${allNonTagQualified.length}/${llmMinNonTagReturn})，停止采样`,
+              );
+              break;
+            }
           }
 
-          // 合格数已达到最小返回阈值，停止所有国家的采样
-          if (allQualified.length >= llmMinReturn) break;
+          // 检查是否两个类型都达到阈值，都达到才停止所有国家采样
+          const tagReachedGlobal = allTagQualified.length >= llmMinTagReturn;
+          const nonTagReachedGlobal = allNonTagQualified.length >= llmMinNonTagReturn;
+          if (tagReachedGlobal && nonTagReachedGlobal) break;
         }
 
-        // 分离 tag 合格和非 tag 合格
-        // tag 任务直接合格（不在 allScores 中），非 tag 任务走 LLM 打分
-        const tagQualified = allQualified.filter(
-          (uid) => !allScores.find((s) => s.uniqueId === uid),
-        );
-        const nonTagQualifiedScores = allScores
-          .filter((s) => s.score >= llmMinScore)
-          .sort((a, b) => b.score - a.score);
-        const nonTagQualified = nonTagQualifiedScores.map((s) => s.uniqueId);
-
+        // 最终合格列表：tag 优先 + 非 tag 按分数排序
         // 限制 tag 占比：最多占 safeLimit 的 70%，留 30% 给非 tag
         const tagMaxCount = Math.floor(safeLimit * 0.7);
-        const tagCount = Math.min(tagQualified.length, tagMaxCount);
+        const tagCount = Math.min(allTagQualified.length, tagMaxCount);
         const nonTagMaxCount = safeLimit - tagCount;
-        const finalNonTagQualified = nonTagQualified.slice(0, nonTagMaxCount);
 
-        // 最终合格列表：tag 优先 + 非 tag 按分数排序
+        const nonTagQualifiedScores = allScores
+          .filter((s) => s.score >= llmMinScore)
+          .sort((a, b) => b.score - a.score);
+        const finalNonTagQualified = nonTagQualifiedScores.slice(0, nonTagMaxCount).map((s) => s.uniqueId);
+
         const qualified = [
-          ...tagQualified.slice(0, tagCount),
+          ...allTagQualified.slice(0, tagCount),
           ...finalNonTagQualified,
         ];