feat: 搜索改为子串匹配，支持多关键词、排除词与分页加载

- 不再依赖 zhparser 分词：关键词按空格拆分，全部需在描述或语音转写中以子串出现；-词表示排除 - 英文/数字关键词要求词首对齐，避免 ai 命中 said - 按描述/转写命中次数打分，同分按发布时间倒序；总数随查询一并返回 - 返回结构化高亮片段（descParts / snippet），前端高亮描述与转写命中位置 - 前端显示描述/语音命中标记、真实结果总数，并支持加载更多 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-08 22:43:15 +08:00
parent 11d9d35002
commit 182ca9cf05
3 changed files with 489 additions and 257 deletions
@@ -1,36 +1,36 @@
 import { NextResponse } from "next/server";
-import { isVideoContentTag } from "@/lib/video-tags";
 import { json } from "@/lib/json";
 import { getFileUrl } from "@/lib/minio";
 import { prisma } from "@/lib/prisma";
+import {
+  buildSnippet,
+  highlightAll,
+  parseSearchQuery,
+  termToPgPattern,
+} from "@/lib/search";
+import { isVideoContentTag } from "@/lib/video-tags";

 type SearchRow = {
-  id: string;
  awemeId: string;
  type: "video" | "image";
-  rank: number;
-  snippet: string;
+  score: number;
+  inDesc: boolean;
+  inTranscript: boolean;
+  rawDesc: string;
+  rawTranscript: string;
  contentTags: string[];
+  total: number;
 };

-function sanitizeSnippetHtml(snippet: string) {
-  return snippet
-    .replaceAll("&", "&amp;")
-    .replaceAll("<", "&lt;")
-    .replaceAll(">", "&gt;")
-    .replaceAll("&lt;mark&gt;", "<mark>")
-    .replaceAll("&lt;/mark&gt;", "</mark>");
-}
-
 export async function GET(req: Request) {
  const { searchParams } = new URL(req.url);
  const q = (searchParams.get("q") || "").trim();
  const rawTag = (searchParams.get("tag") || "").trim();
  const contentTag = rawTag && isVideoContentTag(rawTag) ? rawTag : null;
-  const page = Math.max(1, Number(searchParams.get("page") || 1));
+  const page = Math.max(1, Math.floor(Number(searchParams.get("page")) || 1));
  const limit = Math.min(
    50,
-    Math.max(1, Number(searchParams.get("limit") || 20)),
+    Math.max(1, Math.floor(Number(searchParams.get("limit")) || 20)),
  );
  const offset = (page - 1) * limit;

@@ -38,129 +38,85 @@ export async function GET(req: Request) {
    return NextResponse.json({ error: "Invalid tag" }, { status: 400 });
  }

-  if (!q && !contentTag) {
-    return NextResponse.json({ results: [], total: 0 });
+  const { terms, excludes } = parseSearchQuery(q);
+  if (terms.length === 0 && !contentTag) {
+    return json({ results: [], total: 0, page, limit, terms });
  }

+  // 关键词已在 parseSearchQuery 里转为小写，这里转成正则后与 lower() 后的文本匹配
+  const patterns = terms.map(termToPgPattern);
+  const excludePatterns = excludes.map(termToPgPattern);
+
  try {
-    // 核心查询：Postgres 中文全文检索（视频转写 + 视频desc + 图文desc）+ 相关度排序 + 摘要高亮
-    // 说明：
-    // - 合并搜索来源：VideoTranscript.transcript_zh_tsv、Video.desc 与 ImagePost.desc
-    // - 使用 UNION ALL 合并视频和图文的搜索结果
-    // - rank 取两者匹配得分的最大值，snippet 优先展示转写匹配，否则展示 desc 匹配
-    const rows = q
-      ? await prisma.$queryRaw<SearchRow[]>`
-          WITH tsq AS (
-            SELECT websearch_to_tsquery('zhcfg', ${q}) AS query
+    // 子串匹配：每个关键词都必须出现在「描述 + 语音转写」里（不区分大小写），
+    // 不依赖分词，所以输入任意片段都能搜到。数据量小，全表扫描足够快。
+    // 打分：描述命中权重更高，按出现次数累加（设上限防止长转写刷分）；同分按发布时间倒序。
+    // 没有关键词、只按标签筛选时，score 都是 0，等价于按时间排序。
+    const rows = await prisma.$queryRaw<SearchRow[]>`
+      WITH params AS (
+        SELECT ${patterns}::text[] AS terms, ${excludePatterns}::text[] AS excludes
+      ),
+      docs AS (
+        SELECT
+          v.aweme_id AS "awemeId",
+          'video'::text AS type,
+          v."desc" AS "rawDesc",
+          COALESCE(array_to_string(vt.transcript, E'\n'), '') AS "rawTranscript",
+          v.created_at AS created,
+          v.manual_content_tags AS "contentTags"
+        FROM "Video" v
+        LEFT JOIN "VideoTranscript" vt ON vt."videoId" = v.aweme_id
+        WHERE ${contentTag}::text IS NULL
+           OR v.manual_content_tags @> ARRAY[${contentTag}]::text[]
+
+        UNION ALL
+
+        SELECT
+          ip.aweme_id,
+          'image'::text,
+          ip."desc",
+          '',
+          ip.created_at,
+          ARRAY[]::text[]
+        FROM "ImagePost" ip
+        WHERE ${contentTag}::text IS NULL
+      ),
+      lowered AS (
+        SELECT docs.*, lower(docs."rawDesc") AS d, lower(docs."rawTranscript") AS t
+        FROM docs
+      ),
+      matched AS (
+        SELECT
+          l.*,
+          COALESCE((
+            SELECT SUM(
+              4 * LEAST(regexp_count(l.d, term), 3)
+              + LEAST(regexp_count(l.t, term), 10)
+            )
+            FROM unnest(p.terms) term
+          ), 0)::float8 AS score,
+          EXISTS (SELECT 1 FROM unnest(p.terms) term WHERE l.d ~ term) AS "inDesc",
+          EXISTS (SELECT 1 FROM unnest(p.terms) term WHERE l.t ~ term) AS "inTranscript"
+        FROM lowered l, params p
+        WHERE NOT EXISTS (
+            SELECT 1 FROM unnest(p.terms) term
+            WHERE l.d !~ term AND l.t !~ term
          )
-          -- 视频搜索结果
-          SELECT
-            COALESCE(vt.id, 'desc:' || v.aweme_id) AS id,
-            v.aweme_id AS "awemeId",
-            'video'::text AS type,
-            GREATEST(
-              COALESCE(ts_rank(vt.transcript_zh_tsv, tsq.query), 0),
-              COALESCE(ts_rank(to_tsvector('zhcfg', v.desc), tsq.query), 0)
-            ) AS rank,
-            CASE
-              WHEN vt.transcript_zh_tsv @@ tsq.query THEN
-                ts_headline(
-                  'zhcfg',
-                  array_to_string(vt.transcript, ' '),
-                  tsq.query,
-                  'StartSel=<mark>,StopSel=</mark>,MaxFragments=2,MinWords=2,MaxWords=20'
-                )
-              ELSE
-                ts_headline(
-                  'zhcfg',
-                  v.desc,
-                  tsq.query,
-                  'StartSel=<mark>,StopSel=</mark>,MaxFragments=2,MinWords=2,MaxWords=20'
-                )
-            END AS snippet,
-            v.manual_content_tags AS "contentTags"
-          FROM "Video" v
-          LEFT JOIN "VideoTranscript" vt ON vt."videoId" = v.aweme_id,
-               tsq
-          WHERE
-            (vt.transcript_zh_tsv @@ tsq.query OR to_tsvector('zhcfg', v.desc) @@ tsq.query)
-            AND CASE
-              WHEN ${contentTag}::text IS NULL THEN TRUE
-              ELSE v.manual_content_tags @> ARRAY[${contentTag}]::text[]
-            END
-
-          UNION ALL
-
-          -- 图文搜索结果
-          SELECT
-            'img_desc:' || ip.aweme_id AS id,
-            ip.aweme_id AS "awemeId",
-            'image'::text AS type,
-            ts_rank(to_tsvector('zhcfg', ip.desc), tsq.query) AS rank,
-            ts_headline(
-              'zhcfg',
-              ip.desc,
-              tsq.query,
-              'StartSel=<mark>,StopSel=</mark>,MaxFragments=2,MinWords=2,MaxWords=20'
-            ) AS snippet,
-            ARRAY[]::text[] AS "contentTags"
-          FROM "ImagePost" ip, tsq
-          WHERE to_tsvector('zhcfg', ip.desc) @@ tsq.query
-            AND ${contentTag}::text IS NULL
-
-          ORDER BY rank DESC
-          OFFSET ${offset}
-          LIMIT ${limit};
-        `
-      : await prisma.$queryRaw<SearchRow[]>`
-          SELECT
-            COALESCE(vt.id, 'tag:' || v.aweme_id) AS id,
-            v.aweme_id AS "awemeId",
-            'video'::text AS type,
-            1.0::double precision AS rank,
-            COALESCE(
-              NULLIF(array_to_string(vt.transcript[1:2], ' '), ''),
-              v.desc,
-              '标签匹配'
-            ) AS snippet,
-            v.manual_content_tags AS "contentTags"
-          FROM "Video" v
-          LEFT JOIN "VideoTranscript" vt ON vt."videoId" = v.aweme_id
-          WHERE v.manual_content_tags @> ARRAY[${contentTag}]::text[]
-          ORDER BY v.created_at DESC
-          OFFSET ${offset}
-          LIMIT ${limit};
-        `;
-
-    // 查询总数（参数化，避免注入）
-    const totalRows = q
-      ? await prisma.$queryRaw<{ count: number }[]>`
-          WITH tsq AS (
-            SELECT websearch_to_tsquery('zhcfg', ${q}) AS query
+          AND NOT EXISTS (
+            SELECT 1 FROM unnest(p.excludes) x
+            WHERE l.d ~ x OR l.t ~ x
          )
-          SELECT (
-            (SELECT COUNT(*)::int FROM "Video" v
-             LEFT JOIN "VideoTranscript" vt ON vt."videoId" = v.aweme_id
-             WHERE (vt.transcript_zh_tsv @@ tsq.query OR to_tsvector('zhcfg', v.desc) @@ tsq.query)
-               AND CASE
-                 WHEN ${contentTag}::text IS NULL THEN TRUE
-                 ELSE v.manual_content_tags @> ARRAY[${contentTag}]::text[]
-               END)
-            +
-            (SELECT COUNT(*)::int FROM "ImagePost" ip
-             WHERE to_tsvector('zhcfg', ip.desc) @@ tsq.query
-               AND ${contentTag}::text IS NULL)
-          ) AS count
-          FROM tsq;
-        `
-      : await prisma.$queryRaw<{ count: number }[]>`
-          SELECT COUNT(*)::int AS count
-          FROM "Video" v
-          LEFT JOIN "VideoTranscript" vt ON vt."videoId" = v.aweme_id
-          WHERE v.manual_content_tags @> ARRAY[${contentTag}]::text[];
-        `;
+      )
+      SELECT
+        "awemeId", type, score, "inDesc", "inTranscript",
+        "rawDesc", "rawTranscript", "contentTags",
+        (COUNT(*) OVER ())::int AS total
+      FROM matched
+      ORDER BY score DESC, created DESC, "awemeId"
+      OFFSET ${offset}
+      LIMIT ${limit};
+    `;

-    // 分离视频和图文ID
    const videoIds = rows
      .filter((r) => r.type === "video")
      .map((r) => r.awemeId);
@@ -168,79 +124,92 @@ export async function GET(req: Request) {
      .filter((r) => r.type === "image")
      .map((r) => r.awemeId);

-    // 批量查询视频元信息
-    const videos =
+    const [videos, imagePosts] = await Promise.all([
      videoIds.length > 0
-        ? (
-            await prisma.video.findMany({
-              where: { aweme_id: { in: videoIds } },
-              select: {
-                aweme_id: true,
-                desc: true,
-                cover_url: true,
-                video_url: true,
-                duration_ms: true,
-                author: true,
-              },
-            })
-          ).map((v) => ({
-            ...v,
-            cover_url: getFileUrl(v.cover_url || ""),
-            author: {
-              ...v.author,
-              avatar_url: getFileUrl(v.author.avatar_url || ""),
+        ? prisma.video.findMany({
+            where: { aweme_id: { in: videoIds } },
+            select: {
+              aweme_id: true,
+              desc: true,
+              cover_url: true,
+              video_url: true,
+              duration_ms: true,
+              author: true,
            },
-            video_url: getFileUrl(v.video_url || ""),
-          }))
-        : [];
-
-    // 批量查询图文元信息
-    const imagePosts =
+          })
+        : [],
      imagePostIds.length > 0
-        ? (
-            await prisma.imagePost.findMany({
-              where: { aweme_id: { in: imagePostIds } },
-              select: {
-                aweme_id: true,
-                desc: true,
-                author: true,
-                images: {
-                  orderBy: { order: "asc" },
-                  take: 1,
-                  select: {
-                    url: true,
-                    width: true,
-                    height: true,
-                  },
-                },
+        ? prisma.imagePost.findMany({
+            where: { aweme_id: { in: imagePostIds } },
+            select: {
+              aweme_id: true,
+              desc: true,
+              author: true,
+              images: {
+                orderBy: { order: "asc" },
+                take: 1,
+                select: { url: true, width: true, height: true },
              },
-            })
-          ).map((ip) => ({
-            ...ip,
-            author: {
-              ...ip.author,
-              avatar_url: getFileUrl(ip.author.avatar_url || ""),
            },
-            cover_url: ip.images[0] ? getFileUrl(ip.images[0].url) : null,
-          }))
-        : [];
+          })
+        : [],
+    ]);
+
+    const videoMap = new Map(
+      videos.map((v) => [
+        v.aweme_id,
+        {
+          ...v,
+          cover_url: getFileUrl(v.cover_url || ""),
+          video_url: getFileUrl(v.video_url || ""),
+          author: {
+            ...v.author,
+            avatar_url: getFileUrl(v.author.avatar_url || ""),
+          },
+        },
+      ]),
+    );
+    const imagePostMap = new Map(
+      imagePosts.map((ip) => [
+        ip.aweme_id,
+        {
+          ...ip,
+          author: {
+            ...ip.author,
+            avatar_url: getFileUrl(ip.author.avatar_url || ""),
+          },
+          cover_url: ip.images[0] ? getFileUrl(ip.images[0].url) : null,
+        },
+      ]),
+    );

    return json({
-      results: rows.map((r) => ({
-        ...r,
-        snippet: sanitizeSnippetHtml(r.snippet),
-        video:
-          r.type === "video"
-            ? videos.find((v) => v.aweme_id === r.awemeId)
-            : undefined,
-        imagePost:
-          r.type === "image"
-            ? imagePosts.find((ip) => ip.aweme_id === r.awemeId)
-            : undefined,
-      })),
-      total: totalRows?.[0]?.count ?? 0,
+      results: rows.map((r) => {
+        const matchedIn: ("desc" | "transcript")[] = [];
+        if (r.inDesc) matchedIn.push("desc");
+        if (r.inTranscript) matchedIn.push("transcript");
+        return {
+          id: `${r.type}:${r.awemeId}`,
+          awemeId: r.awemeId,
+          type: r.type,
+          score: r.score,
+          matchedIn,
+          contentTags: r.contentTags,
+          descParts: highlightAll(r.rawDesc, terms),
+          // 转写命中时给出命中片段；只按标签筛选时给出转写开头作为预览
+          snippet:
+            r.inTranscript || (terms.length === 0 && r.rawTranscript)
+              ? buildSnippet(r.rawTranscript, terms)
+              : null,
+          video: r.type === "video" ? videoMap.get(r.awemeId) : undefined,
+          imagePost:
+            r.type === "image" ? imagePostMap.get(r.awemeId) : undefined,
+        };
+      }),
+      total: rows[0]?.total ?? 0,
      page,
      limit,
+      terms,
    });
  } catch (err) {
    console.error("Search error:", err);