diff --git a/docs/queries/douyin-ads-last-7-days.sql b/docs/queries/douyin-ads-last-7-days.sql new file mode 100644 index 0000000..11370c1 --- /dev/null +++ b/docs/queries/douyin-ads-last-7-days.sql @@ -0,0 +1,104 @@ +-- 抖音所有账号:最近七天有没有带「广告」标签的视频 +-- +-- **每个账号一行,一个都不少** —— 包括这几天没发广告的、以及一条作品都没有的号。 +-- 有广告的:那一行带上「最新那条广告视频」的详情;没有的:只剩博主信息,作品列是空的。 +-- +-- 判定口径: +-- * 「最近七天」看的是作品的 **published_at(作者发布时间)**,不是我们什么时候采到的。 +-- 想改成「最近七天新发现的」,把两处 `n.published_at >= ...` 换成 +-- `n3.first_seen_at >= ...` / `n2.first_seen_at >= ...`。 +-- * 广告标签按**名字**匹配(`tg.name = '广告'`)。改了标签名的话这里也要跟着改。 +-- 标签是全局的(不按平台分),而这里只查抖音。 +-- * 窗口下界只在**一个地方**定义(下面那个 `w` 子查询),改天数只改那一处。 +-- 写成 `(UNIX_TIMESTAMP() - 7 * 86400) * 1000` 算的,**不经过时区** —— 库里存的是 +-- 毫秒时间戳,和 MySQL 的 time_zone 设置无关,这样最不容易出错。 +-- * MySQL 5.7 没有窗口函数,挑「最新那条」还是用相关子查询。 +-- +-- 只想看某一个任务:把四处 `t.platform = 'dy'` 换成 `t1.id = 7` 这类条件 +-- (账号清单两处、最新广告视频子查询一处、账号快照子查询一处)。 + +SELECT + acct.creator_hash AS 博主哈希, + ca.alias AS 博主备注, + COALESCE(n.creator_name, st.nickname, '') AS 博主昵称, + st.fans AS 粉丝数, + st.total_favorited AS 总获赞, + st.works_count AS 主页作品数, + -- 这七天里带「广告」标签的视频有几条。0 就是没有 —— 这一行的作品列会是空的。 + (SELECT COUNT(*) + FROM monitor_note n3 + JOIN monitor_task t5 ON t5.id = n3.task_id + WHERE t5.platform = 'dy' + AND n3.creator_hash = acct.creator_hash + AND n3.published_at >= w.since + AND EXISTS (SELECT 1 FROM monitor_note_tag nt + JOIN monitor_tag tg ON tg.id = nt.tag_id + WHERE nt.platform = 'dy' + AND nt.note_id = n3.note_id + AND tg.name = '广告')) AS 近七天广告视频数, + n.note_id AS 作品ID, + n.title AS 作品标题, + n.note_url AS 作品链接, + FROM_UNIXTIME(n.published_at / 1000) AS 发布时间, + m.liked_count AS 点赞, + m.comment_count AS 评论, + m.collected_count AS 收藏, + m.share_count AS 分享, + (SELECT GROUP_CONCAT(t.name ORDER BY t.sort_order SEPARATOR ' / ') + FROM monitor_note_tag nt + JOIN monitor_tag t ON t.id = nt.tag_id + WHERE nt.platform = 'dy' AND nt.note_id = n.note_id) AS 标签, + (SELECT na.alias FROM monitor_note_alias na + WHERE na.platform = 'dy' AND na.note_id = n.note_id LIMIT 1) AS 作品备注 +FROM +-- 七天窗口的下界(毫秒)。**改天数只改这一行。** +(SELECT (UNIX_TIMESTAMP() - 7 * 86400) * 1000 AS since) w +CROSS JOIN ( + -- 账号清单:作品里出现过的 + SELECT n1.creator_hash + FROM monitor_note n1 + JOIN monitor_task t1 ON t1.id = n1.task_id + WHERE t1.platform = 'dy' + UNION + -- ∪ 有账号资料的(包含一条作品都没有的博主) + SELECT s1.creator_hash + FROM monitor_creator_stat s1 + JOIN monitor_task t2 ON t2.id = s1.task_id + WHERE t2.platform = 'dy' +) acct +-- 这七天里最新的一条**带广告标签的**作品。没有就整块为空 —— 但博主那几列照常显示。 +LEFT JOIN monitor_note n + ON n.id = (SELECT n2.id + FROM monitor_note n2 + JOIN monitor_task t3 ON t3.id = n2.task_id + WHERE t3.platform = 'dy' + AND n2.creator_hash = acct.creator_hash + AND n2.published_at >= w.since + AND EXISTS (SELECT 1 FROM monitor_note_tag nt2 + JOIN monitor_tag tg2 ON tg2.id = nt2.tag_id + WHERE nt2.platform = 'dy' + AND nt2.note_id = n2.note_id + AND tg2.name = '广告') + ORDER BY n2.published_at DESC, n2.id DESC + LIMIT 1) +-- 那条作品最近一轮的指标 +LEFT JOIN monitor_note_metric m + ON m.id = (SELECT m2.id + FROM monitor_note_metric m2 + WHERE m2.task_id = n.task_id + AND m2.note_id = n.note_id + ORDER BY m2.run_id DESC + LIMIT 1) +LEFT JOIN monitor_creator_alias ca + ON ca.platform = 'dy' AND ca.creator_hash = acct.creator_hash +-- 博主最近一轮的账号级指标 +LEFT JOIN monitor_creator_stat st + ON st.id = (SELECT s2.id + FROM monitor_creator_stat s2 + JOIN monitor_task t4 ON t4.id = s2.task_id + WHERE t4.platform = 'dy' + AND s2.creator_hash = acct.creator_hash + ORDER BY s2.run_id DESC + LIMIT 1) +-- 有广告的排前面(数量降序),其余按粉丝数 +ORDER BY 近七天广告视频数 DESC, st.fans DESC, acct.creator_hash