社交媒体行为分析报告
Q1. 跨表一致性检查
| 平台 | 用户均值 | 平台表值 | 差值 |
|---|---|---|---|
| Bluesky | 142.4854 | 8.0000 | 134.4854 |
| Discord | 146.5087 | 40.0000 | 106.5087 |
| 140.2221 | 33.0000 | 107.2221 | |
| 140.0627 | 35.0000 | 105.0627 | |
| 140.3275 | 11.0000 | 129.3275 | |
| 141.9667 | 14.0000 | 127.9667 | |
| RedNote | 141.2082 | 25.0000 | 116.2082 |
| 141.7874 | 24.0000 | 117.7874 | |
| Snapchat | 141.2541 | 30.0000 | 111.2541 |
| Telegram | 142.3301 | 18.0000 | 124.3301 |
| Threads | 137.7528 | 12.0000 | 125.7528 |
| TikTok | 142.5701 | 58.0000 | 84.5701 |
| 143.4046 | 28.0000 | 115.4046 | |
| X (Twitter) | 140.6729 | 22.0000 | 118.6729 |
| YouTube | 141.1732 | 48.0000 | 93.1732 |
Pearson 相关系数: 0.3143
绝对差最大的 3 个平台:
- Bluesky: 绝对差 = 134.4854
- LinkedIn: 绝对差 = 129.3275
- Pinterest: 绝对差 = 127.9667
Q2. 独立性检验
2×2 列联表:
| has_purchased_via_social=False | has_purchased_via_social=True | |
|---|---|---|
| is_content_creator=False | 13048 | 8160 |
| is_content_creator=True | 2340 | 1452 |
chi2 = 0.0390, dof = 1, p_value = 0.8435
结论: 在 α=0.05 水平上，无法拒绝独立性假设
Q3. 逻辑回归：预测是否通过社交平台购买过
Train AUC: 0.5220 Test AUC: 0.5006 Test Accuracy: 0.6156
系数绝对值 Top 5 的特征:
| 特征 | 系数 |
|---|---|
| primary_platform_Threads | 0.1585 |
| income_bracket_$150K+ | -0.1203 |
| primary_platform_WhatsApp | 0.1056 |
| primary_platform_TikTok | 0.1035 |
| primary_platform_Snapchat | 0.1025 |
Q4. Simpson’s 风险检查
r_all (整体): -0.3698 r_creator (创作者): -0.3615 r_non_creator (非创作者): -0.3713
符号反转或量级减半以上检查:
- 未检测到符号反转或量级减半以上
Q5. K-Means 行为聚类
每簇规模:
| 簇 | 用户数 | 占比(%) |
|---|---|---|
| 0 | 6585 | 26.3400 |
| 1 | 7136 | 28.5440 |
| 2 | 2658 | 10.6320 |
| 3 | 8621 | 34.4840 |
每簇原始特征均值:
| 簇 | daily_screen_time_minutes | engagement_rate_pct | posts_per_week | addiction_level_1_to_10 |
|---|---|---|---|---|
| 0 | 116.6691 | 2.9098 | 2.1001 | 2.1072 |
| 1 | 207.9937 | 1.7783 | 2.2190 | 4.8065 |
| 2 | 141.5662 | 1.8286 | 10.5779 | 3.0899 |
| 3 | 104.3116 | 1.1635 | 2.0220 | 1.8397 |
业务标签:
- 簇 0: 高参与观众
- 簇 1: 中度普通用户
- 簇 2: 重度创作者
- 簇 3: 低活跃用户
Q6. 异常参与度用户
残差绝对值 Top 5 用户:
| user_id | followers_count | engagement_rate_pct | predicted | residual |
|---|---|---|---|---|
| USR-020194 | 1025 | 6.1200 | 1.8388 | 4.2812 |
| USR-020577 | 1733 | 5.9400 | 1.8455 | 4.0945 |
| USR-010877 | 1581 | 5.7100 | 1.8443 | 3.8657 |
| USR-005491 | 1324 | 5.6000 | 1.8420 | 3.7580 |
| USR-001540 | 5706 | 5.5300 | 1.8608 | 3.6692 |