2nd/10_Wiki/Topics/Architecture/사용자_제작_콘텐츠(UGC).md

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

id	title	category	status	canonical_id	aliases	duplicate_of	source_trust_level	confidence_score	verification_status	tags	raw_sources	last_reinforced	github_commit	tech_stack
wiki-2026-0508-사용자-제작-콘텐츠-ugc	사용자 제작 콘텐츠 (UGC)	10_Wiki/Topics	verified	self	User-Generated Content UGC User Content	none	A	0.9	applied	ugc content moderation platform architecture		2026-05-10	pending	language framework typescript nextjs

사용자 제작 콘텐츠 (UGC)

매 한 줄

"매 user 의 platform 에서 create 하는 content". 매 Web 2.0 (2004) 의 core concept — Wikipedia, YouTube, Reddit 의 fuel. 2026 년 매 LLM-generated content + human content 의 mix 의 challenge — 매 provenance, moderation, copyright 의 critical.

매 핵심

매 lifecycle

1. Creation: 매 upload, post, comment.
2. Storage: 매 object storage (S3, R2) + DB metadata.
3. Moderation: 매 automated (Perspective API, OpenAI Moderation) + human review.
4. Distribution: 매 CDN + recommendation algorithm.
5. Lifecycle end: 매 delete, archive, takedown (DMCA).

매 challenges (2026)

• AI-generated content: 매 deepfake, slop — provenance (C2PA) 의 필요.
• Moderation scale: 매 LLM moderator 의 cost-effective.
• Copyright: 매 training data 의 lawsuit (NYT v OpenAI 등).
• Liability: 매 Section 230 (US), DSA (EU) 의 compliance.

매 응용

1. Social media — 매 post, video, comment.
2. Reviews/ratings — 매 e-commerce.
3. Forums — 매 Reddit, Discord, HN.
4. Wiki — 매 collaborative knowledge.

💻 패턴

Upload + S3 presigned URL

// API: generate presigned URL
import { S3Client, PutObjectCommand } from '@aws-sdk/client-s3';
import { getSignedUrl } from '@aws-sdk/s3-request-presigner';

export async function POST(req: Request) {
  const { userId } = await authenticate(req);
  const { filename, contentType } = await req.json();

  if (!ALLOWED_TYPES.includes(contentType)) {
    return Response.json({ error: 'invalid type' }, { status: 400 });
  }

  const key = `uploads/${userId}/${crypto.randomUUID()}-${filename}`;
  const url = await getSignedUrl(
    s3,
    new PutObjectCommand({ Bucket: 'ugc', Key: key, ContentType: contentType }),
    { expiresIn: 300 }
  );

  return Response.json({ url, key });
}

Moderation pipeline (text)

import OpenAI from 'openai';
const openai = new OpenAI();

async function moderateText(text: string) {
  const result = await openai.moderations.create({
    model: 'omni-moderation-latest',
    input: text,
  });
  const r = result.results[0];
  if (r.flagged) {
    return { allowed: false, categories: r.categories };
  }
  return { allowed: true };
}

Image moderation (vision)

// 매 NSFW + violence detection — Claude Opus 4.7 vision.
import Anthropic from '@anthropic-ai/sdk';
const client = new Anthropic();

async function moderateImage(imageUrl: string) {
  const res = await client.messages.create({
    model: 'claude-opus-4-7',
    max_tokens: 200,
    messages: [{
      role: 'user',
      content: [
        { type: 'image', source: { type: 'url', url: imageUrl } },
        { type: 'text', text: 'Classify this image: SFW, NSFW, violence, hate, none. JSON only.' },
      ],
    }],
  });
  return JSON.parse(res.content[0].text);
}

Provenance (C2PA)

// 매 image 의 origin metadata — AI 생성 여부 의 verify.
import { read } from 'c2pa';

async function checkProvenance(imageUrl: string) {
  const c2pa = await read(imageUrl);
  if (!c2pa) return { verified: false };

  const manifest = c2pa.manifestStore?.activeManifest;
  return {
    verified: c2pa.manifestStore?.validationStatus === 'valid',
    generator: manifest?.claimGenerator,
    aiGenerated: manifest?.assertions.some((a) => a.label.startsWith('c2pa.actions.ai_')),
  };
}

Vote + ranking

// 매 Reddit-style — Wilson score interval.
function wilsonScore(up: number, down: number) {
  const n = up + down;
  if (n === 0) return 0;
  const z = 1.96; // 95% confidence
  const p = up / n;
  return (p + z*z/(2*n) - z * Math.sqrt((p*(1-p) + z*z/(4*n))/n)) / (1 + z*z/n);
}

const posts = await db.post.findMany();
posts.sort((a, b) => wilsonScore(b.up, b.down) - wilsonScore(a.up, a.down));

Rate limit (anti-spam)

import { Ratelimit } from '@upstash/ratelimit';
import { Redis } from '@upstash/redis';

const ratelimit = new Ratelimit({
  redis: Redis.fromEnv(),
  limiter: Ratelimit.slidingWindow(10, '1 h'), // 10 posts/hour
});

const { success } = await ratelimit.limit(`post:${userId}`);
if (!success) throw new Error('rate limit');

DMCA takedown flow

async function handleDmcaTakedown(claim: DmcaClaim) {
  // 1. Hide content immediately
  await db.content.update({
    where: { id: claim.contentId },
    data: { status: 'TAKEDOWN_PENDING' },
  });

  // 2. Notify uploader (counter-notice option)
  await notifyUploader(claim);

  // 3. Log for audit
  await db.dmcaLog.create({ data: { ...claim, processedAt: new Date() } });
}

매 결정 기준

Content type	Moderation
Text comment	OpenAI Moderation API + human review
Image	Claude vision + perceptual hash (NCMEC PhotoDNA)
Video	Frame sampling + ML + community flag
Long-form (article)	Pre-publish review + post-flag
Realtime chat	Profanity filter + ML in pipeline

기본값: 매 LLM moderation + human escalation queue.

🔗 Graph

• Adjacent: Content Moderation · C2PA

🤖 LLM 활용

언제: user-content platform 의 moderation, ranking, provenance 의 design 시. 언제 X: closed editorial content (news site) — 매 UGC 무관.

❌ 안티패턴

• No moderation: 매 spam/abuse 의 platform 의 destroy.
• Heavy pre-moderation only: 매 latency + cost — 매 hybrid 의 better.
• Naive ranking (order by votes desc): 매 first-mover advantage — Wilson score 의 better.
• Ignore copyright: 매 DMCA 의 ignore — legal liability.

🧪 검증 / 중복

• Verified (OpenAI Moderation docs; C2PA spec 2.0; OWASP UGC guidelines; DSA 2024).
• 신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — UGC lifecycle + moderation + C2PA