analysis_02_timing.py

#!/usr/bin/env python3
"""
川普密碼 分析 #2 — 發文時間規律
什麼時候發？頻率突變前後發生了什麼？
"""

import json
from collections import Counter, defaultdict
from datetime import datetime, timezone
from pathlib import Path
from statistics import median
from utils import est_hour

BASE = Path(__file__).parent
DATA = BASE / "data"


def main():
    with open(BASE / "clean_president.json", 'r', encoding='utf-8') as f:
        posts = json.load(f)

    originals = [p for p in posts if p['has_text'] and not p['is_retweet']]

    print("=" * 70)
    print("⏰ 分析 #2: 發文時間規律")
    print(f"   分析對象: 就任後原創貼文 {len(originals)} 篇")
    print("=" * 70)

    # --- 1. 每小時分布（UTC → EST 轉換，Trump 在東岸） ---
    hour_dist = Counter()
    hour_by_month = defaultdict(Counter)

    for p in originals:
        est_h, _ = est_hour(p['created_at'])
        hour_dist[est_h] += 1
        month = p['created_at'][:7]
        hour_by_month[month][est_h] += 1

    print(f"\n🕐 發文時段分布 (美東時間 EST):")
    print("-" * 60)
    max_count = max(hour_dist.values())
    for h in range(24):
        count = hour_dist.get(h, 0)
        bar = '█' * int(count / max_count * 40) if max_count > 0 else ''
        period = "🌙" if h < 6 else ("☀️" if h < 12 else ("🌅" if h < 18 else "🌙"))
        print(f"  {h:02d}:00 {period} {count:4d} {bar}")

    # 深夜推文 (12am-5am EST)
    night_posts = [p for p in originals if est_hour(p['created_at'])[0] < 5]
    print(f"\n🌙 深夜推文 (12am-5am EST): {len(night_posts)} 篇 ({len(night_posts)/len(originals)*100:.1f}%)")
    if night_posts:
        print("   最近 5 篇深夜推文:")
        for p in night_posts[:5]:
            est_h, est_m = est_hour(p['created_at'])
            print(f"   {p['created_at'][:16]} (EST {est_h}:{est_m:02d}) | {p['content'][:80]}...")

    # --- 2. 每日發文量 ---
    print(f"\n📅 每日發文量分布:")
    print("-" * 60)
    daily = Counter()
    for p in originals:
        daily[p['created_at'][:10]] += 1

    counts = sorted(daily.values())
    avg_daily = sum(counts) / len(counts)
    # P4-1: 空值防護 + P3-8: 中位數改用 statistics.median
    median_count = median(counts) if counts else 0
    print(f"  平均每天: {avg_daily:.1f} 篇")
    if counts:
        print(f"  最少: {counts[0]} 篇")
        print(f"  最多: {counts[-1]} 篇")
    print(f"  中位數: {median_count} 篇")

    # Top 10 最多發文的日子
    print(f"\n🔥 Top 10 發文最多的日子:")
    for date, count in daily.most_common(10):
        # 那天的第一篇看看主題
        day_posts = [p for p in originals if p['created_at'][:10] == date]
        topic = day_posts[0]['content'][:60] if day_posts else ''
        bar = '█' * count
        print(f"  {date} | {count:3d}篇 | {bar} | {topic}...")

    # 沉默日（0 篇或極少篇）
    print(f"\n🤫 沉默日（≤2 篇）:")
    all_dates = sorted(daily.keys())
    quiet_days = [(d, daily[d]) for d in all_dates if daily[d] <= 2]
    print(f"  總計 {len(quiet_days)} 天")
    for d, c in quiet_days[-10:]:
        print(f"  {d} | {c} 篇")

    # --- 3. 星期分布 ---
    print(f"\n📊 星期分布:")
    print("-" * 60)
    weekday_dist = Counter()
    weekday_names = ['一', '二', '三', '四', '五', '六', '日']
    for p in originals:
        dt = datetime.fromisoformat(p['created_at'].replace('Z', '+00:00'))
        weekday_dist[dt.weekday()] += 1

    for wd in range(7):
        count = weekday_dist.get(wd, 0)
        bar = '█' * int(count / max(weekday_dist.values()) * 30)
        print(f"  週{weekday_names[wd]} | {count:4d} {bar}")

    # --- 4. 發文間隔分析 ---
    print(f"\n⏱️ 發文間隔分析:")
    print("-" * 60)
    intervals = []
    sorted_posts = sorted(originals, key=lambda p: p['created_at'])
    for i in range(1, len(sorted_posts)):
        dt1 = datetime.fromisoformat(sorted_posts[i-1]['created_at'].replace('Z', '+00:00'))
        dt2 = datetime.fromisoformat(sorted_posts[i]['created_at'].replace('Z', '+00:00'))
        diff_minutes = (dt2 - dt1).total_seconds() / 60
        intervals.append({
            'minutes': diff_minutes,
            'date': sorted_posts[i]['created_at'][:16],
            'content': sorted_posts[i]['content'][:60]
        })

    intervals_min = sorted([i['minutes'] for i in intervals])
    # P4-1: 空值防護
    if not intervals_min:
        print("  無間隔資料")
        return

    print(f"  最短間隔: {intervals_min[0]:.0f} 分鐘")
    print(f"  最長間隔: {intervals_min[-1]:.0f} 分鐘 ({intervals_min[-1]/60:.1f} 小時)")
    print(f"  平均間隔: {sum(intervals_min)/len(intervals_min):.0f} 分鐘")
    # P3-8: 用 statistics.median
    print(f"  中位數間隔: {median(intervals_min):.0f} 分鐘")

    # 連續轟炸（5分鐘內連發多篇）
    bursts = [i for i in intervals if i['minutes'] < 5]
    print(f"\n🔥 連續轟炸（5分鐘內連發）: {len(bursts)} 次")

    # 長沉默後的第一篇（沉默 > 12小時）
    long_silence = [i for i in intervals if i['minutes'] > 720]
    print(f"\n😶 長時間沉默（>12小時）後的第一篇: {len(long_silence)} 次")
    for item in long_silence[:10]:
        hours = item['minutes'] / 60
        print(f"  沉默 {hours:.1f}h → {item['date']} | {item['content']}...")

    # --- 5. 發文量趨勢（每週） ---
    print(f"\n📈 每週發文量趨勢:")
    print("-" * 60)
    weekly = defaultdict(int)
    for p in originals:
        dt = datetime.fromisoformat(p['created_at'].replace('Z', '+00:00'))
        # ISO week
        year, week, _ = dt.isocalendar()
        key = f"{year}-W{week:02d}"
        weekly[key] += 1

    weeks = sorted(weekly.keys())
    for w in weeks[-16:]:  # 最近 16 週
        count = weekly[w]
        bar = '█' * (count // 2)
        print(f"  {w} | {count:3d} {bar}")

    # 存結果
    results = {
        'hourly_distribution_est': dict(hour_dist),
        'daily_counts': dict(daily.most_common()),
        'weekday_distribution': {weekday_names[k]: v for k, v in weekday_dist.items()},
        'night_posts_count': len(night_posts),
        'burst_count': len(bursts),
        'long_silence_count': len(long_silence),
        'avg_daily': round(avg_daily, 1),
        'avg_interval_minutes': round(sum(intervals_min)/len(intervals_min), 0),
    }
    with open(DATA / 'results_02_timing.json', 'w', encoding='utf-8') as f:
        json.dump(results, f, ensure_ascii=False, indent=2)

    print(f"\n💾 詳細結果存入 results_02_timing.json")


if __name__ == '__main__':
    main()