douyin-archive/app/api/search/route.ts

import { NextResponse } from "next/server";
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 = {
  awemeId: string;
  type: "video" | "image";
  score: number;
  inDesc: boolean;
  inTranscript: boolean;
  rawDesc: string;
  rawTranscript: string;
  contentTags: string[];
  total: number;
};

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, Math.floor(Number(searchParams.get("page")) || 1));
  const limit = Math.min(
    50,
    Math.max(1, Math.floor(Number(searchParams.get("limit")) || 20)),
  );
  const offset = (page - 1) * limit;

  if (rawTag && !contentTag) {
    return NextResponse.json({ error: "Invalid tag" }, { status: 400 });
  }

  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 {
    // 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
          )
          AND NOT EXISTS (
            SELECT 1 FROM unnest(p.excludes) x
            WHERE l.d ~ x OR l.t ~ x
          )
      )
      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};
    `;

    const videoIds = rows
      .filter((r) => r.type === "video")
      .map((r) => r.awemeId);
    const imagePostIds = rows
      .filter((r) => r.type === "image")
      .map((r) => r.awemeId);

    const [videos, imagePosts] = await Promise.all([
      videoIds.length > 0
        ? 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,
            },
          })
        : [],
      imagePostIds.length > 0
        ? 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 },
              },
            },
          })
        : [],
    ]);

    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) => {
        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);
    return NextResponse.json({ error: "Search failed" }, { status: 500 });
  }
}