freedom_lime | 作者
掘金 | 来源
https://juejin.im/post/6844904202590748679#heading-0
4. 用户行为转化漏斗
# 导入相关包
from pyecharts import options as opts
from pyecharts.charts import Funnel
import mathbehavior['action_time'] = pd.to_datetime(behavior['action_time'],format ='%Y-%m-%d %H:%M:%S')# 用户整体行为分布
type_dis = behavior['type'].value_counts().reset_index()
type_dis['rate'] = round((type_dis['type'] / type_dis['type'].sum()),3)type_dis.style.bar(color='skyblue',subset=['rate'])用户整体行为中，有82.6%行为为浏览，实际支付操作仅占6.4,除此之外，用户评论及收藏的行为占比也较低，应当增强网站有用户之间的互动，提高评论数量和收藏率。
df_con = behavior[['user_id', 'sku_id', 'action_time', 'type']]df_pv = df_con[df_con['type'] == 'pv']
df_fav = df_con[df_con['type'] == 'fav']
df_cart = df_con[df_con['type'] == 'cart']
df_pay = df_con[df_con['type'] == 'pay']
df_pv_uid = df_con[df_con['type'] == 'pv']['user_id'].unique()
df_fav_uid = df_con[df_con['type'] == 'fav']['user_id'].unique()
df_cart_uid = df_con[df_con['type'] == 'cart']['user_id'].unique()
df_pay_uid = df_con[df_con['type'] == 'pay']['user_id'].unique()pv - buy
fav_cart_list = set(df_fav_uid) | set(df_cart_uid)pv_pay_df = pd.merge(left=df_pv, right=df_pay, how='inner', on=['user_id', 'sku_id'],
suffixes=('_pv', '_pay'))pv_pay_df = pv_pay_df[(~pv_pay_df['user_id'].isin(fav_cart_list)) & (pv_pay_df['action_time_pv'] < pv_pay_df['action_time_pay'])]uv = behavior['user_id'].nunique()
pv_pay_num = pv_pay_df['user_id'].nunique()
pv_pay_data = pd.DataFrame({'type':['浏览','付款'],'num':[uv,pv_pay_num]})
pv_pay_data['conversion_rates'] = (round((pv_pay_data['num'] / pv_pay_data['num'][0]),4) * 100)attr1 = list(pv_pay_data.type)
values1 = list(pv_pay_data.conversion_rates)
data1 = [[attr1[i], values1[i]] for i in range(len(attr1))]# 用户行为转化漏斗可视化
pv_pay=(Funnel(opts.InitOpts(width="600px", height="300px"))
.add(
series_name="",
data_pair=data1,
gap=2,
tooltip_opts=opts.TooltipOpts(trigger="item", formatter="{b} : {c}%"),
label_opts=opts.LabelOpts(is_show=True, position="inside"),
itemstyle_opts=opts.ItemStyleOpts(border_color="#fff", border_width=1)
)
.set_global_opts(title_opts=opts.TitleOpts(title="用户行为转化漏斗图"))
)
pv_pay.render_notebook()pv - cart - pay
pv_cart_df = pd.merge(left=df_pv, right=df_cart, how='inner', on=['user_id', 'sku_id'],
suffixes=('_pv', '_cart'))pv_cart_df = pv_cart_df[pv_cart_df['action_time_pv'] < pv_cart_df['action_time_cart']]
pv_cart_df = pv_cart_df[~pv_cart_df['user_id'].isin(df_fav_uid)]pv_cart_pay_df = pd.merge(left=pv_cart_df, right=df_pay, how='inner', on=['user_id', 'sku_id'])pv_cart_pay_df = pv_cart_pay_df[pv_cart_pay_df['action_time_cart'] < pv_cart_pay_df['action_time']]uv = behavior['user_id'].nunique()
pv_cart_num = pv_cart_df['user_id'].nunique()
pv_cart_pay_num = pv_cart_pay_df['user_id'].nunique()
pv_cart_pay_data = pd.DataFrame({'type':['浏览','加购','付款'],'num':[uv,pv_cart_num,pv_cart_pay_num]})
pv_cart_pay_data['conversion_rates'] = (round((pv_cart_pay_data['num'] / pv_cart_pay_data['num'][0]),4) * 100)attr2 = list(pv_cart_pay_data.type)
values2 = list(pv_cart_pay_data.conversion_rates)
data2 = [[attr2[i], values2[i]] for i in range(len(attr2))]# 用户行为转化漏斗可视化
pv_cart_buy=(Funnel(opts.InitOpts(width="600px", height="300px"))
.add(
series_name="",
data_pair=data2,
gap=2,
tooltip_opts=opts.TooltipOpts(trigger="item", formatter="{b} : {c}%"),
label_opts=opts.LabelOpts(is_show=True, position="inside"),
itemstyle_opts=opts.ItemStyleOpts(border_color="#fff", border_width=1)
)
.set_global_opts(title_opts=opts.TitleOpts(title="用户行为转化漏斗图"))
)
pv_cart_buy.render_notebook()pv - fav - pay
pv_fav_df = pd.merge(left=df_pv, right=df_fav, how='inner', on=['user_id', 'sku_id'],
suffixes=('_pv', '_fav'))pv_fav_df = pv_fav_df[pv_fav_df['action_time_pv'] < pv_fav_df['action_time_fav']]
pv_fav_df = pv_fav_df[~pv_fav_df['user_id'].isin(df_cart_uid)]pv_fav_pay_df = pd.merge(left=pv_fav_df, right=df_pay, how='inner', on=['user_id', 'sku_id'])pv_fav_pay_df = pv_fav_pay_df[pv_fav_pay_df['action_time_fav'] < pv_fav_pay_df['action_time']]uv = behavior['user_id'].nunique()
pv_fav_num = pv_fav_df['user_id'].nunique()
pv_fav_pay_num = pv_fav_pay_df['user_id'].nunique()
pv_fav_pay_data = pd.DataFrame({'type':['浏览','收藏','付款'],'num':[uv,pv_fav_num,pv_fav_pay_num]})
pv_fav_pay_data['conversion_rates'] = (round((pv_fav_pay_data['num'] / pv_fav_pay_data['num'][0]),4) * 100)attr3 = list(pv_fav_pay_data.type)
values3 = list(pv_fav_pay_data.conversion_rates)
data3 = [[attr3[i], values3[i]] for i in range(len(attr3))]# 用户行为转化漏斗可视化
pv_fav_buy=(Funnel(opts.InitOpts(width="600px", height="300px"))
.add(
series_name="",
data_pair=data3,
gap=2,
tooltip_opts=opts.TooltipOpts(trigger="item", formatter="{b} : {c}%"),
label_opts=opts.LabelOpts(is_show=True, position="inside"),
itemstyle_opts=opts.ItemStyleOpts(border_color="#fff", border_width=1)
)
.set_global_opts(title_opts=opts.TitleOpts(title="用户行为转化漏斗图"))
)
pv_fav_buy.render_notebook()pv - fav - cart - pay
pv_fav = pd.merge(left=df_pv, right=df_fav, how='inner', on=['user_id', 'sku_id'],
suffixes=('_pv', '_fav'))pv_fav = pv_fav[pv_fav['action_time_pv'] < pv_fav['action_time_fav']]pv_fav_cart = pd.merge(left=pv_fav, right=df_cart, how='inner', on=['user_id', 'sku_id'])pv_fav_cart = pv_fav_cart[pv_fav_cart['action_time_fav']<pv_fav_cart['action_time']]pv_fav_cart_pay = pd.merge(left=pv_fav_cart, right=df_pay, how='inner', on=['user_id', 'sku_id'],
suffixes=('_cart', '_pay'))pv_fav_cart_pay = pv_fav_cart_pay[pv_fav_cart_pay['action_time_cart']<pv_fav_cart_pay['action_time_pay']]uv = behavior['user_id'].nunique()
pv_fav_n = pv_fav['user_id'].nunique()
pv_fav_cart_n = pv_fav_cart['user_id'].nunique()
pv_fav_cart_pay_n = pv_fav_cart_pay['user_id'].nunique()
pv_fav_cart_pay_data = pd.DataFrame({'type':['浏览','收藏','加购','付款'],'num':[uv,pv_fav_n,pv_fav_cart_n,pv_fav_cart_pay_n]})
pv_fav_cart_pay_data['conversion_rates'] = (round((pv_fav_cart_pay_data['num'] / pv_fav_cart_pay_data['num'][0]),4) * 100)attr4 = list(pv_fav_cart_pay_data.type)
values4 = list(pv_fav_cart_pay_data.conversion_rates)
data4 = [[attr4[i], values4[i]] for i in range(len(attr4))]# 用户行为转化漏斗可视化
pv_fav_buy=(Funnel(opts.InitOpts(width="600px", height="300px"))
.add(
series_name="",
data_pair=data4,
gap=2,
tooltip_opts=opts.TooltipOpts(trigger="item", formatter="{b} : {c}%"),
label_opts=opts.LabelOpts(is_show=True, position="inside"),
itemstyle_opts=opts.ItemStyleOpts(border_color="#fff", border_width=1)
)
.set_global_opts(title_opts=opts.TitleOpts(title="用户行为转化漏斗图"))
)
pv_fav_buy.render_notebook()不同路径用户消费时间间隔分析：
pv - cart - pay
pcp_interval = pv_cart_pay_df.groupby(['user_id', 'sku_id']).apply(lambda x: (x.action_time.min() - x.action_time_cart.min())).reset_index()pcp_interval['interval'] = pcp_interval[0].apply(lambda x: x.seconds) / 3600
pcp_interval['interval'] = pcp_interval['interval'].apply(lambda x: math.ceil(x))fig, ax = plt.subplots(figsize=[16,6])
sns.countplot(pcp_interval['interval'],palette='Set1')
for p in ax.patches:
ax.annotate('{:.2f}%'.format(100*p.get_height()/len(pcp_interval['interval'])), (p.get_x() + 0.1, p.get_height() + 100))
ax.set_yscale("log")
plt.title('pv-cart-pay路径用户消费时间间隔')pv - fav - pay
pfp_interval = pv_fav_pay_df.groupby(['user_id', 'sku_id']).apply(lambda x: (x.action_time.min() - x.action_time_fav.min())).reset_index()pfp_interval['interval'] = pfp_interval[0].apply(lambda x: x.seconds) / 3600
pfp_interval['interval'] = pfp_interval['interval'].apply(lambda x: math.ceil(x))fig, ax = plt.subplots(figsize=[16,6])
sns.countplot(pfp_interval['interval'],palette='Set1')
for p in ax.patches:
ax.annotate('{:.2f}%'.format(100*p.get_height()/len(pfp_interval['interval'])), (p.get_x() + 0.1, p.get_height() + 10))
ax.set_yscale("log")
plt.title('pv-fav-pay路径用户消费时间间隔')两种路径下大部分用户均在4小时内完成了支付，大部分用户的购物意向很明确，也侧面说明了网站的商品分类布局和购物结算方式比较合理。
# SQL
# 漏斗图
SELECT type, COUNT(DISTINCT user_id) user_num
FROM behavior_sql
GROUP BY type
ORDER BY COUNT(DISTINCT user_id) DESC
SELECT COUNT(DISTINCT b.user_id) AS pv_fav_num,COUNT(DISTINCT c.user_id) AS pv_fav_pay_num
FROM
((SELECT DISTINCT user_id, sku_id, action_time FROM users WHERE type='pv' ) AS a
LEFT JOIN
(SELECT DISTINCT user_id, sku_id, action_time FROM users WHERE type='fav'
AND user_id NOT IN
(SELECT DISTINCT user_id
FROM behavior_sql
WHERE type = 'cart')) AS b
ON a.user_id = b.user_id AND a.sku_id = b.sku_id AND a.action_time <= b.action_time
LEFT JOIN
(SELECT DISTINCT user_id,sku_id,item_category,times_new FROM users WHERE behavior_type='pay') AS c
ON b.user_id = c.user_id AND b.sku_id = c.sku_id AND AND b.action_time <= c.action_time)
;比较四种不同的转化方式，最有效的转化路径为浏览直接付款转化率为21.46%，其次为浏览加购付款，转化率为12.47%，可以发现随着结算方式越来越复杂转化率越来越低。
加购的方式比收藏购买的方式转化率要高，推其原因为购物车接口进入方便且可以做不同商家比价用，而收藏则需要更繁琐的操作才可以查看到商品，因此转化率较低。
可以优化商品搜索功能，提高商品搜索准确度、易用性，减少用户搜索时间。
根据用户喜好在首页进行商品推荐，优化重排商品详情展示页，提高顾客下单欲望，提供一键购物等简化购物步骤的功能。
客服也可以留意加购及关注用户，适时推出优惠福利及时解答用户问题，引导用户购买以进一步提高转化率。
对于用户消费时间间隔，可以通过限时领券购买、限时特惠价格等进一步缩短用户付款时间，提高订单量。
5. 用户留存率分析
#留存率
first_day = datetime.date(datetime.strptime('2018-03-30', '%Y-%m-%d'))
fifth_day = datetime.date(datetime.strptime('2018-04-03', '%Y-%m-%d'))
tenth_day = datetime.date(datetime.strptime('2018-04-08', '%Y-%m-%d'))
fifteenth_day = datetime.date(datetime.strptime('2018-04-13', '%Y-%m-%d'))
#第一天新用户数
user_num_first = behavior[behavior['date'] == first_day]['user_id'].to_frame()
#第五天留存用户数
user_num_fifth = behavior[behavior['date'] == fifth_day ]['user_id'].to_frame()
#第十留存用户数
user_num_tenth = behavior[behavior['date'] == tenth_day]['user_id'].to_frame()
#第十五天留存用户数
user_num_fifteenth = behavior[behavior['date'] == fifteenth_day]['user_id'].to_frame()#第五天留存率
fifth_day_retention_rate = round((pd.merge(user_num_first, user_num_fifth).nunique())
/ (user_num_first.nunique()),4).user_id
#第十天留存率
tenth_day_retention_rate = round((pd.merge(user_num_first, user_num_tenth ).nunique())
/ (user_num_first.nunique()),4).user_id
#第十五天留存率
fifteenth_day_retention_rate = round((pd.merge(user_num_first, user_num_fifteenth).nunique())
/ (user_num_first.nunique()),4).user_idretention_rate = pd.DataFrame({'n日后留存率':['第五天留存率','第十天留存率','第十五天留存率'],
'Rate':[fifth_day_retention_rate,tenth_day_retention_rate,fifteenth_day_retention_rate]})# 留存率可视化
fig, ax = plt.subplots(figsize=[16,6])
sns.barplot(x='n日后留存率', y='Rate', data=retention_rate,
palette='Set1')
x=list(range(0,3))
for a,b in zip(x,retention_rate['Rate']):
plt.text(a, b + 0.001, '%.2f%%' % (b*100), ha='center', va= 'bottom',fontsize=12)
plt.title('用户留存率')留存率反应了产品质量和保留用户的能力，按照Facebook平台流传出留存率“40–20–10”规则（规则中的数字表示的是次日留存率、第7日留存率和第30日留存率），统计周期内第五日留存率为22.81%，第15日留存率为17.44%。
反映出平台的用户依赖性较高，也因平台发展已经到达稳定阶段，用户保留率不会发生较大波动，数据量足够的情况下可以以年为单位，计算按月的留存率。要合理安排消息推送，推出签到有奖等机制提高用户粘性，进一步提高留存率。
# SQL
#n日后留存率=(注册后的n日后还登录的用户数)/第一天新增总用户数
create table retention_rate as select count(distinct user_id) as user_num_first from behavior_sql
where date = '2018-03-30';
alter table retention_rate add column user_num_fifth INTEGER;
update retention_rate set user_num_fifth=
(select count(distinct user_id) from behavior_sql
where date = '2018-04-03' and user_id in (SELECT user_id FROM behavior_sql
WHERE date = '2018-03-30'));
alter table retention_rate add column user_num_tenth INTEGER;
update retention_rate set user_num_tenth=
(select count(distinct user_id) from behavior_sql
where date = '2018-04-08' and user_id in (SELECT user_id FROM behavior_sql
WHERE date = '2018-03-30'));
alter table retention_rate add column user_num_fifteenth INTEGER;
update retention_rate set user_num_fifteenth=
(select count(distinct user_id) from behavior_sql
where date = '2018-04-13' and user_id in (SELECT user_id FROM behavior_sql
WHERE date = '2018-03-30'));
SELECT CONCAT(ROUND(100*user_num_fifth/user_num_first,2),'%')AS fifth_day_retention_rate,
CONCAT(ROUND(100*user_num_tenth/user_num_first,2),'%')AS tenth_day_retention_rate,
CONCAT(ROUND(100*user_num_fifteenth/user_num_first,2),'%')AS fifteenth_day_retention_rate
from retention_rate;6. 商品销量分析
# 商品总数
behavior['sku_id'].nunique()# OUTPUT
239007# 商品被购前产生平均操作次数
sku_df = behavior[behavior['sku_id'].isin(behavior[behavior['type'] == 'pay']['sku_id'].unique())].groupby('sku_id')['type'].value_counts().unstack(fill_value=0)
sku_df['total'] = sku_df.sum(axis=1)
sku_df['avg_beha'] = round((sku_df['total'] / sku_df['pay']), 2)fig, ax = plt.subplots(figsize=[8,6])
sns.scatterplot(x='avg_beha', y='pay', data=sku_df, palette='Set1')
ax.set_xscale("log")
ax.set_yscale("log")
plt.xlabel('平均操作次数')
plt.ylabel('销量')左下角操作少购买少，属于冷门购买频率较低的产品。
左上角操作少购买多，属于快消类产品，可选择品牌少，少数品牌垄断的行业。
右下角操作多购买少，品牌多，但是购买频率低，应为贵重物品类。
右上角操作多购买多，大众品牌，可选多，被购买频次高。
# 商品销量排行
sku_num = (behavior[behavior['type'] == 'pay'].groupby('sku_id')['type'].count().to_frame()
.rename(columns={'type':'total'}).reset_index())
# 销量大于1000的商品
topsku = sku_num[sku_num['total'] > 1000].sort_values(by='total',ascending=False)
# 单个用户共购买商品种数
sku_num_per_user = (behavior[behavior['type'] == 'pay']).groupby(['user_id'])['sku_id'].nunique()topsku.set_index('sku_id').style.bar(color='skyblue',subset=['total'])在计算周期内订单数均突破1000以上的共有13款产品，其中152092这款产品订单数最高为1736. 推出商品组合有优惠等，提高单个用户购买产品种数。
# SQL
# sku销量排行
SELECT sku_id, COUNT(type) sku_num FROM behavior_sql
WHERE type = 'pay'
GROUP BY sku_id
HAVING sku_num > 1000
ORDER BY sku_num DESC;7. RFM用户分层
#RFM
#由于缺少M（金额）列，仅通过R（最近一次购买时间）和F（消费频率）对用户进行价值分析
buy_group = behavior[behavior['type']=='pay'].groupby('user_id')['date']#将2018-04-13作为每个用户最后一次购买时间来处理
final_day = datetime.date(datetime.strptime('2018-04-14', '%Y-%m-%d'))#最近一次购物时间
recent_buy_time = buy_group.apply(lambda x:final_day-x.max())
recent_buy_time = recent_buy_time.reset_index().rename(columns={'date':'recent'})
recent_buy_time['recent'] = recent_buy_time['recent'].map(lambda x:x.days)#近十五天内购物频率
buy_freq = buy_group.count().reset_index().rename(columns={'date':'freq'})RFM = pd.merge(recent_buy_time,buy_freq,on='user_id')RFM['R'] = pd.qcut(RFM.recent,2,labels=['1','0'])
#天数小标签为1天数大标签为0
RFM['F'] = pd.qcut(RFM.freq.rank(method='first'),2,labels=['0','1'])
#频率大标签为1频率小标签为0RFM['RFM'] = RFM['R'].astype(int).map(str) + RFM['F'].astype(int).map(str)dict_n={'01':'重要保持客户',
'11':'重要价值客户',
'10':'重要挽留客户',
'00':'一般发展客户'}#用户标签
RFM['用户等级'] = RFM['RFM'].map(dict_n)RFM_pie = RFM['用户等级'].value_counts().reset_index()
RFM_pie['Rate'] = RFM_pie['用户等级'] / RFM_pie['用户等级'].sum()fig, ax = plt.subplots(figsize=[16,6])
plt.pie(RFM_pie['Rate'], labels = RFM_pie['index'], startangle = 90,autopct="%1.2f%%",
counterclock = False,colors = ['yellowgreen', 'gold', 'lightskyblue', 'lightcoral'])
plt.axis('square')
plt.title('RFM用户分层')不同类型用户占比差异较小，应提升重要价值各户的占比，减小一般发展客户的占比。
通过RFM模型对用户价值进行分类，对不同价值用户应采取不同的运营策略：
对于重要价值客户来说，要提高该部分用户的满意度，服务升级，发放特别福利，增大该部分用户留存率，在做运营推广时也要给与特别关注，避免引起用户反感。
对于重要保持客户，他们购物频次较高，但最近一段时间没有消费，可以推送相关其他商品，发放优惠卷、赠品和促销信息等，唤回该部分用户。
对于重要挽留客户，他们最近消费过，但购物频次较低，可以通过问卷有礼的方式找出其对平台的不满，提升购物体验，增大用户粘性。
对于一般发展客户，做到定期发送邮件或短信唤回，努力将其转化为重要保持客户或重要挽留客户。
# SQL
# RFM
CREATE VIEW RF_table AS
SELECT user_id, DATEDIFF('2018-04-14',MAX(date)) AS R_days,
COUNT(*) AS F_count
FROM behavior_sql WHERE type='pay' GROUP BY user_id;
SELECT AVG(R_days), AVG(F_count)
FROM RF_table
create view RF_ layer as
SELECT user_id, (CASE WHEN R_days < 7.1697 THEN 1 ELSE 0 END) AS R,
(CASE WHEN F_count < 1.2129 THEN 0 ELSE 1 END) AS F
FROM RF_table
ORDER BY user_id DESC;
create view customer_value as
select user_id, R, F, (CASE WHEN R=1 and F=1 THEN "重要价值客户"
WHEN R=1 and F=0 THEN "重要挽留客户"
WHEN R=0 and F=1 THEN "重要保持客户"
WHEN R=0 and F=0 THEN "一般发展客户" ELSE 0 END) as 用户价值
FROM RF_ layer;
SELECT * FROM customer_value;5
总结
1.可以增加渠道推广投入，进行精准人群推广，推出新用户福利，吸引新用户，推出团购、分享有礼等活动促进老带新，推出促销活动刺激老用户，提高访客数和浏览量。提高产品质量，提高商品详情页对用户的吸引力，降低跳失率。
2.根据用户操作随时间变化规律来开展营销活动，使活动更容易触达用户，在用户访问高峰期多推送用户感兴趣商品。
3.复购率较低，说明用户对平台购物体验不满，需要找出用户槽点，提高用户购物满意度，优化商品推送机制，对老用户给予特别福利，提高他们所享受权益。转化率也偏低，需要改善平台搜索机制降低提高搜索效率，优化购物路径降低购物复杂度，改善商品详情信息展示方式便于信息的获取。
4.留存率相对稳定，为进一步提高留存率，可以定期推出秒杀活动，推出专享优惠券，推出签到有礼环节，增加用户浏览时长和深度，提高用户粘性。分析用户对产品的真实使用感受与评价，提高用户忠诚度。
5.通过RFM对用户进行分层，将用户从一个整体拆分成特征明显的群体，有针对性的采取不同的营销方法进行精准化营销，用有限的公司资源优先服务于公司最重要的客户。
- END -
本文为转载分享&推荐阅读，若侵权请联系后台删除
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
