跳转至

高级联动与窗口分析

目标

在基础仪表盘之外,使用 v4.1 的组合粒度表做更细的联动分析,回答“某个日期窗口、某个类目、某类用户分层的行为是否不同”。

建议仪表盘名称:

LZ-组6_高级联动与窗口分析

使用限制

date_window_category_segment_metrics 有 137 万行,适合按条件聚合,不建议全表明细展示。

推荐只做:

  • 日期窗口趋势
  • 类目 Top N
  • 分层对比
  • 条件聚合

不要直接把 137 万行全部放表格里。

图表1:连续日期窗口整体趋势

保存名称:

趋势-连续日期窗口行为

可视化:Line

SELECT
  CONCAT(window_start, ' 至 ', window_end) AS `日期窗口`,
  window_days AS `窗口天数`,
  SUM(pv_count) AS `浏览量`,
  SUM(cart_count) AS `加购量`,
  SUM(buy_count) AS `购买行为数`,
  SUM(buy_count) / NULLIF(SUM(pv_count), 0) AS `购买行为率`
FROM date_window_category_segment_metrics
GROUP BY window_start, window_end, window_days
ORDER BY window_start, window_end;

图表2:指定窗口类目购买 Top20

保存名称:

排行-窗口类目购买Top20

可视化:Bar

变量:

  • start_date:Date,显示名称可填“开始日期”
  • end_date:Date,显示名称可填“结束日期”
SELECT
  category_id AS `类目ID`,
  SUM(pv_count) AS `浏览量`,
  SUM(buy_count) AS `购买行为数`,
  SUM(pv_buy_users) / NULLIF(SUM(pv_users), 0) AS `浏览到购买用户转化率`
FROM date_window_category_segment_metrics
WHERE window_start = {{start_date}}
  AND window_end = {{end_date}}
GROUP BY category_id
ORDER BY `购买行为数` DESC
LIMIT 20;

默认可测试:

start_date:2017-11-25
end_date:2017-12-03

图表3:指定类目的用户分层趋势

保存名称:

趋势-指定类目分层行为

可视化:Line

变量:

  • category_id_param:Number,显示名称可填“类目ID”
SELECT
  date AS `日期`,
  CASE segment
    WHEN 'non_buyer' THEN '未购买用户'
    WHEN 'single_day_buyer' THEN '单日购买用户'
    WHEN 'repeat_day_buyer' THEN '跨日复购用户'
    WHEN 'high_frequency_buyer' THEN '高频购买用户'
    ELSE segment
  END AS `用户分层`,
  pv_count AS `浏览量`,
  cart_count AS `加购量`,
  buy_count AS `购买行为数`,
  pv_to_buy_user_rate AS `浏览到购买用户转化率`
FROM daily_category_segment_metrics
WHERE category_id = {{category_id_param}}
ORDER BY date, segment;

图表4:分层购买贡献

保存名称:

柱状-分层购买贡献

可视化:Bar

SELECT
  CASE segment
    WHEN 'non_buyer' THEN '未购买用户'
    WHEN 'single_day_buyer' THEN '单日购买用户'
    WHEN 'repeat_day_buyer' THEN '跨日复购用户'
    WHEN 'high_frequency_buyer' THEN '高频购买用户'
    ELSE segment
  END AS `用户分层`,
  SUM(pv_count) AS `浏览量`,
  SUM(cart_count) AS `加购量`,
  SUM(buy_count) AS `购买行为数`,
  SUM(pv_buy_users) / NULLIF(SUM(pv_users), 0) AS `浏览到购买用户转化率`
FROM daily_category_segment_metrics
GROUP BY segment
ORDER BY `购买行为数` DESC;

图表5:高转化类目与用户分层组合

保存名称:

表格-高转化类目分层组合

可视化:Table

SELECT
  category_id AS `类目ID`,
  CASE segment
    WHEN 'non_buyer' THEN '未购买用户'
    WHEN 'single_day_buyer' THEN '单日购买用户'
    WHEN 'repeat_day_buyer' THEN '跨日复购用户'
    WHEN 'high_frequency_buyer' THEN '高频购买用户'
    ELSE segment
  END AS `用户分层`,
  SUM(pv_users) AS `浏览用户数`,
  SUM(buy_users) AS `购买用户数`,
  SUM(pv_buy_users) / NULLIF(SUM(pv_users), 0) AS `浏览到购买用户转化率`,
  SUM(pv_count) AS `浏览量`,
  SUM(buy_count) AS `购买行为数`
FROM daily_category_segment_metrics
GROUP BY category_id, segment
HAVING SUM(pv_users) >= 1000
ORDER BY `浏览到购买用户转化率` DESC
LIMIT 100;

适合答辩的讲法

基础看板解决“整体发生了什么”,高级联动看板解决“在哪个窗口、哪个类目、哪类用户上更明显”。为了避免 UV 等不可加指标被错误拼接,v4.1 直接生成了日期、类目、用户分层组合粒度表,而不是在看板端临时拼表。