Comment construire un scoreur de qualité d'audience TikTok

import os, requests, json, statistics
from datetime import datetime

API_KEY = os.environ['SCAVIO_API_KEY']
TH = {'Authorization': f'Bearer {API_KEY}', 'Content-Type': 'application/json'}

def get_creator_stats(username):
    data = requests.post('https://api.scavio.dev/api/v1/tiktok/user/videos',
        headers=TH, json={'username': username}).json()
    videos = data.get('videos', data.get('data', {}).get('videos', []))
    stats = []
    for v in videos:
        s = v.get('stats', {})
        plays = s.get('playCount', 0)
        likes = s.get('diggCount', 0)
        comments = s.get('commentCount', 0)
        shares = s.get('shareCount', 0)
        if plays > 0:
            stats.append({
                'plays': plays,
                'likes': likes,
                'comments': comments,
                'shares': shares,
                'like_rate': likes / plays,
                'comment_rate': comments / plays,
                'share_rate': shares / plays,
                'engagement_rate': (likes + comments + shares) / plays,
            })
    return stats

CREATORS = ['charlidamelio', 'khaby.lame', 'addisonre']

creator_data = {}
for username in CREATORS:
    stats = get_creator_stats(username)
    creator_data[username] = stats
    if stats:
        avg_plays = statistics.mean(s['plays'] for s in stats)
        avg_er = statistics.mean(s['engagement_rate'] for s in stats)
        print(f'  @{username:20} | {len(stats)} videos | Avg plays: {avg_plays:,.0f} | Avg ER: {avg_er*100:.1f}%')
print(f'\nCost: ${len(CREATORS) * 0.005:.3f}')

def score_audience_quality(username, stats):
    if not stats or len(stats) < 3:
        return {'username': username, 'score': 0, 'reason': 'Insufficient data'}
    signals = {}
    # 1. Engagement consistency (real audiences have some variance, bots are too consistent)
    er_values = [s['engagement_rate'] for s in stats]
    er_cv = statistics.stdev(er_values) / statistics.mean(er_values) if statistics.mean(er_values) > 0 else 0
    # Healthy CV is 0.3-0.8. Too low = bot-like, too high = bought views
    if 0.2 <= er_cv <= 1.0:
        signals['consistency'] = 25
    elif er_cv < 0.2:
        signals['consistency'] = 5  # Suspiciously consistent
    else:
        signals['consistency'] = 10  # Too erratic
    # 2. Like-to-view ratio (healthy: 3-15%)
    avg_like_rate = statistics.mean(s['like_rate'] for s in stats)
    if 0.03 <= avg_like_rate <= 0.15:
        signals['like_ratio'] = 25
    elif avg_like_rate < 0.01:
        signals['like_ratio'] = 5  # Views but no likes = bought views
    else:
        signals['like_ratio'] = 15
    # 3. Comment-to-like ratio (healthy: 1-5%)
    avg_comment_to_like = statistics.mean(s['comments'] / s['likes'] if s['likes'] > 0 else 0 for s in stats)
    if 0.01 <= avg_comment_to_like <= 0.05:
        signals['comment_quality'] = 25
    elif avg_comment_to_like < 0.005:
        signals['comment_quality'] = 5  # Likes but no comments = suspicious
    else:
        signals['comment_quality'] = 15
    # 4. Share rate (real engagement drives shares)
    avg_share_rate = statistics.mean(s['share_rate'] for s in stats)
    signals['share_quality'] = min(25, int(avg_share_rate * 2500))
    total = sum(signals.values())
    return {
        'username': username,
        'score': total,
        'signals': signals,
        'avg_er': statistics.mean(er_values),
        'er_cv': er_cv,
        'avg_like_rate': avg_like_rate,
    }

print(f'\n=== Audience Quality Scores ===')
scores = []
for username, stats in creator_data.items():
    result = score_audience_quality(username, stats)
    scores.append(result)
    grade = 'A' if result['score'] >= 80 else 'B' if result['score'] >= 60 else 'C' if result['score'] >= 40 else 'F'
    print(f'  @{username:20} | Score: {result["score"]:3}/100 | Grade: {grade}')
    if result.get('signals'):
        for signal, value in result['signals'].items():
            print(f'    {signal:20} {value:3}/25')

def audience_quality_report(scores):
    print(f'\n{"=" * 60}')
    print(f'  TIKTOK AUDIENCE QUALITY REPORT')
    print(f'  Date: {datetime.now().strftime("%Y-%m-%d")}')
    print(f'{"=" * 60}')
    scores.sort(key=lambda x: x['score'], reverse=True)
    for i, s in enumerate(scores, 1):
        grade = 'A' if s['score'] >= 80 else 'B' if s['score'] >= 60 else 'C' if s['score'] >= 40 else 'F'
        status = 'RECOMMENDED' if grade in ['A', 'B'] else 'CAUTION' if grade == 'C' else 'AVOID'
        print(f'\n  {i}. @{s["username"]} - Grade {grade} ({s["score"]}/100) - {status}')
        if s.get('avg_er'):
            print(f'     Engagement Rate: {s["avg_er"]*100:.1f}%')
            print(f'     Engagement Variance: {s.get("er_cv", 0):.2f} (0.3-0.8 is healthy)')
            print(f'     Like Rate: {s.get("avg_like_rate", 0)*100:.1f}% (3-15% is healthy)')
    # Summary
    recommended = sum(1 for s in scores if s['score'] >= 60)
    caution = sum(1 for s in scores if 40 <= s['score'] < 60)
    avoid = sum(1 for s in scores if s['score'] < 40)
    print(f'\n  Summary:')
    print(f'    Recommended: {recommended}')
    print(f'    Caution: {caution}')
    print(f'    Avoid: {avoid}')
    print(f'\n  Cost: ${len(scores) * 0.005:.3f}')
    print(f'  vs. HypeAuditor: $299/mo for audience quality reports')

audience_quality_report(scores)