feat: hashtag 列表显示按共现计算的相关 tag

- 新增 /api/hashtag/related:从作品描述实时解析 hashtag,按「同一作品里一起出现」计算相关 tag
- 打分取子话题(候选多与目标同现)与上位话题(目标多与热门候选同现)两者较大值,
  过滤 cos、二次元等万金油 tag 与平台活动、纯数字 tag
- 播放页侧栏的 hashtag 列表顶部显示相关 tag,点击在侧栏内预览(不打断当前播放),
  在预览里点作品才切换到该 tag 的列表

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-11-09 01:18:04 +08:00
co-authored by Claude Opus 5.5
parent b84fcec77b
commit 2391abdeea
7 changed files with 279 additions and 7 deletions
+28
View File
@@ -0,0 +1,28 @@
import { NextResponse } from "next/server";
import { normalizeHashtag } from "@/lib/hashtag";
import { relatedHashtags } from "@/lib/hashtag-related.server";
import { json } from "@/lib/json";
// 相关 hashtag(按共现计算):GET /api/hashtag/related?name=丰川祥子
export async function GET(req: Request) {
const { searchParams } = new URL(req.url);
const name = normalizeHashtag(searchParams.get("name") || "");
if (!name) {
return NextResponse.json({ error: "Invalid hashtag" }, { status: 400 });
}
const limit = Math.min(
30,
Math.max(1, Math.floor(Number(searchParams.get("limit")) || 12)),
);
try {
return json(await relatedHashtags(name, limit));
} catch (err) {
console.error("Related hashtags error:", err);
return NextResponse.json(
{ error: "Related hashtags failed" },
{ status: 500 },
);
}
}
export const dynamic = "force-dynamic";
+10
View File
@@ -471,6 +471,14 @@ export default function AwemeDetailClient({
return `/search?${params.toString()}`; return `/search?${params.toString()}`;
}, [listKind, contextQuery]); }, [listKind, contextQuery]);
const hashtagName = useMemo(
() =>
listKind === "hashtag"
? new URLSearchParams(contextQuery).get("hashtag")
: null,
[listKind, contextQuery],
);
const listVisible = panel.open && activeTab === "list"; const listVisible = panel.open && activeTab === "list";
const worksVisible = panel.open && activeTab === "works"; const worksVisible = panel.open && activeTab === "works";
@@ -976,11 +984,13 @@ export default function AwemeDetailClient({
count: listContext.total, count: listContext.total,
content: ( content: (
<ContextListTab <ContextListTab
key={contextQuery}
active={panel.open && activeTab === "list"} active={panel.open && activeTab === "list"}
context={listContext} context={listContext}
list={contextList} list={contextList}
currentId={data.aweme_id} currentId={data.aweme_id}
contextQuery={contextQuery} contextQuery={contextQuery}
hashtagName={hashtagName}
searchPageHref={searchPageHref} searchPageHref={searchPageHref}
onSelect={handleWorkSelect} onSelect={handleWorkSelect}
/> />
@@ -1,7 +1,9 @@
import { Hash, Search } from "lucide-react"; import { ArrowLeft, Hash, Search } from "lucide-react";
import Link from "next/link"; import Link from "next/link";
import { memo } from "react"; import { memo, useState } from "react";
import type { PlaylistState } from "../hooks/usePlaylist"; import { playbackContextQuery } from "@/lib/playback-context";
import { type PlaylistState, usePlaylist } from "../hooks/usePlaylist";
import { useRelatedHashtags } from "../hooks/useRelatedHashtags";
import type { PlaybackContextInfo } from "../types.ts"; import type { PlaybackContextInfo } from "../types.ts";
import { WorksGrid } from "./WorksGrid"; import { WorksGrid } from "./WorksGrid";
@@ -11,23 +13,62 @@ interface ContextListTabProps {
list: PlaylistState; list: PlaylistState;
currentId: string; currentId: string;
contextQuery: string; contextQuery: string;
/** hashtag 上下文时的 tag 名(用来取相关 tag) */
hashtagName: string | null;
/** 对应的搜索结果页链接(搜索上下文才有) */ /** 对应的搜索结果页链接(搜索上下文才有) */
searchPageHref: string | null; searchPageHref: string | null;
onSelect: () => void; onSelect: () => void;
} }
// 侧栏的「搜索结果 / #hashtag / 标签」列表:和上下切换的顺序完全一致 // 侧栏的「搜索结果 / #hashtag / 标签」列表:和上下切换的顺序完全一致。
// hashtag 列表上方显示相关 tag(按共现计算);点相关 tag 只在侧栏里预览它的作品,不打断当前播放,
// 在预览里点作品才切换到那个 tag 的列表。
export const ContextListTab = memo(function ContextListTab({ export const ContextListTab = memo(function ContextListTab({
active, active,
context, context,
list, list,
currentId, currentId,
contextQuery, contextQuery,
hashtagName,
searchPageHref, searchPageHref,
onSelect, onSelect,
}: ContextListTabProps) { }: ContextListTabProps) {
// 正在预览的相关 tag(null 表示显示当前列表本身)
const [browsing, setBrowsing] = useState<string | null>(null);
const viewTag = browsing ?? hashtagName;
const related = useRelatedHashtags(viewTag, active);
const browseQuery = browsing
? playbackContextQuery({ kind: "hashtag", name: browsing })
: "";
const browseList = usePlaylist(browseQuery, active && Boolean(browsing));
const pickTag = (name: string) => {
setBrowsing(
name.toLowerCase() === hashtagName?.toLowerCase() ? null : name,
);
};
const Icon = context.kind === "hashtag" ? Hash : Search; const Icon = context.kind === "hashtag" ? Hash : Search;
const header = ( const header = browsing ? (
<>
<button
type="button"
onClick={() => setBrowsing(null)}
className="grid h-8 w-8 shrink-0 place-items-center rounded-full bg-white/10 text-white/80 hover:bg-white/20 cursor-pointer"
aria-label={`返回 ${context.label}`}
title={`返回 ${context.label}`}
>
<ArrowLeft size={16} />
</button>
<div className="min-w-0 flex-1">
<div className="truncate font-semibold text-white">#{browsing}</div>
<div className="text-xs text-white/50">
{related ? `共 ${related.count} 个 · ` : ""}点作品切换到这个 tag
</div>
</div>
</>
) : (
<> <>
<Icon size={18} className="shrink-0 text-rose-300" /> <Icon size={18} className="shrink-0 text-rose-300" />
<div className="min-w-0 flex-1"> <div className="min-w-0 flex-1">
@@ -47,14 +88,34 @@ export const ContextListTab = memo(function ContextListTab({
</> </>
); );
const relatedRow =
viewTag && related && related.related.length > 0 ? (
<div className="flex items-center gap-1.5 overflow-x-auto px-3 py-2 border-b border-white/10 comment-scroll">
<span className="shrink-0 text-xs text-white/40">相关</span>
{related.related.map((tag) => (
<button
key={tag.name}
type="button"
onClick={() => pickTag(tag.name)}
className="shrink-0 rounded-full border border-white/15 bg-white/5 px-2.5 py-1 text-xs text-white/80 hover:border-white/30 hover:text-white cursor-pointer"
title={`一起出现在 ${tag.together} 个作品里,#${tag.name} 共 ${tag.count} 个作品`}
>
#{tag.name}
<span className="ml-1 text-white/40">{tag.together}</span>
</button>
))}
</div>
) : null;
return ( return (
<WorksGrid <WorksGrid
active={active} active={active}
list={list} list={browsing ? browseList : list}
currentId={currentId} currentId={currentId}
hrefQuery={contextQuery} hrefQuery={browsing ? browseQuery : contextQuery}
onSelect={onSelect} onSelect={onSelect}
header={header} header={header}
subheader={relatedRow}
/> />
); );
}); });
@@ -16,6 +16,8 @@ interface WorksGridProps {
singleHint?: string; singleHint?: string;
/** 网格上方的标题栏 */ /** 网格上方的标题栏 */
header: ReactNode; header: ReactNode;
/** 标题栏下方的附加内容(如相关 tag) */
subheader?: ReactNode;
} }
function formatCount(n: number): string { function formatCount(n: number): string {
@@ -46,6 +48,7 @@ export const WorksGrid = memo(function WorksGrid({
onSelect, onSelect,
singleHint, singleHint,
header, header,
subheader,
}: WorksGridProps) { }: WorksGridProps) {
const scrollRef = useRef<HTMLDivElement>(null); const scrollRef = useRef<HTMLDivElement>(null);
const currentRef = useRef<HTMLAnchorElement>(null); const currentRef = useRef<HTMLAnchorElement>(null);
@@ -74,6 +77,7 @@ export const WorksGrid = memo(function WorksGrid({
<div className="flex items-center gap-3 px-3 py-3 border-b border-white/10 shrink-0"> <div className="flex items-center gap-3 px-3 py-3 border-b border-white/10 shrink-0">
{header} {header}
</div> </div>
{subheader}
<div ref={scrollRef} className="comment-scroll flex-1 overflow-auto p-2"> <div ref={scrollRef} className="comment-scroll flex-1 overflow-auto p-2">
{list.loading && !list.items && ( {list.loading && !list.items && (
@@ -0,0 +1,37 @@
import { useEffect, useState } from "react";
import type { RelatedHashtagsResult } from "@/lib/hashtag-related.server";
const cache = new Map<string, RelatedHashtagsResult>();
/** 取某个 hashtag 的相关 tag(按共现计算);name 为空或 enabled 为 false 时不请求。 */
export function useRelatedHashtags(
name: string | null,
enabled: boolean,
): RelatedHashtagsResult | null {
const key = name?.toLowerCase() ?? "";
const [result, setResult] = useState<RelatedHashtagsResult | null>(
() => cache.get(key) ?? null,
);
useEffect(() => {
if (!key || !enabled) return;
const cached = cache.get(key);
setResult(cached ?? null);
if (cached) return;
let cancelled = false;
fetch(`/api/hashtag/related?name=${encodeURIComponent(key)}`)
.then((res) => (res.ok ? res.json() : null))
.then((data: RelatedHashtagsResult | null) => {
if (!data) return;
cache.set(key, data);
if (!cancelled) setResult(data);
})
.catch(() => {});
return () => {
cancelled = true;
};
}, [key, enabled]);
return result?.name === key ? result : null;
}
+114
View File
@@ -0,0 +1,114 @@
import {
HASHTAG_EXTRACT_PG,
META_HASHTAG_PG,
META_HASHTAGS,
} from "@/lib/hashtag";
import { prisma } from "@/lib/prisma";
export type RelatedHashtag = {
name: string;
/** 和目标 tag 一起出现的作品数 */
together: number;
/** 这个 tag 自己的作品数 */
count: number;
};
export type RelatedHashtagsResult = {
name: string;
count: number;
related: RelatedHashtag[];
};
const CACHE_MS = 60_000;
const cache = new Map<string, { at: number; value: RelatedHashtagsResult }>();
/**
* 按「同一个作品描述里一起出现」计算相关 hashtag(共现图),每次实时从描述里解析,不需要额外的表。
*
* 打分取两个方向里较大的那个:
* - 子话题:候选 tag 大多和目标一起出现(together / 候选数),乘 ln(1+together) 防止只出现一次的巧合排到前面;
* 例:#明日方舟 → #明日方舟七周年、#特蕾西娅
* - 上位话题:目标大多和候选一起出现(together / 目标数),乘 ln(1+候选数),让热门的上位 tag 排前面;
* 例:#圣聆初雪(只有 2 个作品)→ #明日方舟
* cos、二次元这类「万金油」tag 与平台活动 tag 不参与推荐。
*/
export async function relatedHashtags(
rawName: string,
limit = 12,
): Promise<RelatedHashtagsResult> {
const name = rawName.toLowerCase();
const key = `${name}:${limit}`;
const hit = cache.get(key);
if (hit && Date.now() - hit.at < CACHE_MS) return hit.value;
const rows = await prisma.$queryRaw<
{ name: string; together: number; count: number; target: number }[]
>`
WITH ht AS (
SELECT DISTINCT w.aweme_id, lower(m[1]) AS tag
FROM (
SELECT aweme_id, "desc" FROM "Video"
UNION ALL
SELECT aweme_id, "desc" FROM "ImagePost"
) w,
regexp_matches(w."desc", ${HASHTAG_EXTRACT_PG}, 'g') m
),
target AS (SELECT aweme_id FROM ht WHERE tag = ${name}),
target_count AS (SELECT count(*)::int AS c FROM target),
co AS (
SELECT ht.tag, count(*)::int AS together
FROM ht JOIN target USING (aweme_id)
WHERE ht.tag <> ${name}
GROUP BY ht.tag
),
counts AS (
SELECT tag, count(*)::int AS c FROM ht
WHERE tag IN (SELECT tag FROM co)
GROUP BY tag
)
SELECT
co.tag AS name,
co.together,
counts.c AS count,
tc.c AS target
FROM co
JOIN counts USING (tag)
CROSS JOIN target_count tc
WHERE co.together >= CASE WHEN tc.c >= 4 THEN 2 ELSE 1 END
AND co.tag <> ALL (${META_HASHTAGS}::text[])
AND co.tag !~ ${META_HASHTAG_PG}
ORDER BY GREATEST(
co.together::float8 / counts.c * ln(1 + co.together),
co.together::float8 / tc.c * ln(1 + counts.c)
) DESC, co.together DESC, co.tag
LIMIT ${limit}
`;
let count = rows[0]?.target;
if (count === undefined) {
// 没有相关 tag 时单独数一下目标 tag 的作品数
const [{ c }] = await prisma.$queryRaw<{ c: number }[]>`
SELECT count(DISTINCT w.aweme_id)::int AS c
FROM (
SELECT aweme_id, "desc" FROM "Video"
UNION ALL
SELECT aweme_id, "desc" FROM "ImagePost"
) w,
regexp_matches(w."desc", ${HASHTAG_EXTRACT_PG}, 'g') m
WHERE lower(m[1]) = ${name}
`;
count = c;
}
const value: RelatedHashtagsResult = {
name,
count,
related: rows.map(({ name, together, count }) => ({
name,
together,
count,
})),
};
cache.set(key, { at: Date.now(), value });
return value;
}
+18
View File
@@ -47,3 +47,21 @@ export function hashtagToPgPattern(name: string): string {
const escaped = name.toLowerCase().replace(/[.*+?^${}()|[\]\\]/g, "\\$&"); const escaped = name.toLowerCase().replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
return `#${escaped}([\\s${STOP_CHARS}]|$)`; return `#${escaped}([\\s${STOP_CHARS}]|$)`;
} }
/** 从描述里提取全部 hashtag 名的 Postgres 正则(ARE),配合 regexp_matches(..., 'g') 使用 */
export const HASHTAG_EXTRACT_PG = `#([^\\s${STOP_CHARS}]+)`;
// 「万金油」tag:几乎和什么都一起出现,不提供话题信息,不作为相关 tag 推荐
export const META_HASHTAGS = [
"cos",
"cosplay",
"coser",
"二次元",
"ooc致歉",
"搞笑",
"抽象",
"vlog日常",
"内容过于真实",
];
/** 平台活动类 tag(xx计划、xx挑战赛…)和纯数字 tag 的 Postgres 正则 */
export const META_HASHTAG_PG = "(计划|挑战赛|大赛|扶持|热门|浪潮)$|^[0-9]+$";