【Hive】（十四）Hive 项目实战之电子商务消费行为分析

一、前言

Hive 学习过程中的一个练习项目，如果不妥的地方或者更好的建议，欢迎指出！
我们主要进行一下一些练习：

• 数据结构
• 数据清洗
• 基于Hive的数据分析

二、项目需求

首先和大家讲一下这个项目的需求：

对某零售企业最近1年门店收集的数据进行数据分析

• 潜在客户画像
• 用户消费统计
• 门店的资源利用率
• 消费的特征人群定位
• 数据的可视化展现

三、数据结构

本次练习一共用到四张表，如下：

Customer表

Transaction表

Store表

Review表

四、项目实战

Create HDFS Folder

hdfs dfs -mkdir -p /tmp/shopping/data/customer
hdfs dfs -mkdir -p /tmp/shopping/data/transaction
hdfs dfs -mkdir -p /tmp/shopping/data/store
hdfs dfs -mkdir -p /tmp/shopping/data/review

Upload the file to HDFS

hdfs dfs -put /opt/soft/data/customer_details.csv /tmp/shopping/data/customer/
hdfs dfs -put /opt/soft/data/transaction_details.csv /tmp/shopping/data/transaction/
hdfs dfs -put /opt/soft/data/store_details.csv /tmp/shopping/data/store/
hdfs dfs -put /opt/soft/data/store_review.csv /tmp/shopping/data/review/

Create database

drop database if exists shopping cascade
create database shopping

Use database

use shopping

Create external table

创建四张对应的外部表，也就是本次项目中的近源表。

create external table if not exists ext_customer_details(
customer_id string,
first_name string,
last_name string,
email string,
gender string,
address string,
country string,
language string,
job string,
credit_type string,
credit_no string
)
row format delimited fields terminated by ','
location '/tmp/shopping/data/customer/'
tblproperties('skip.header.line.count'='1')

create external table if not exists ext_transaction_details(
transaction_id string,
customer_id string,
store_id string,
price double,
product string,
buydate string,
buytime string
)
row format delimited fields terminated by ','
location '/tmp/shopping/data/transaction'
tblproperties('skip.header.line.count'='1')

create external table if not exists ext_store_details(
store_id string,
store_name string,
employee_number int
)
row format delimited fields terminated by ','
location '/tmp/shopping/data/store/'
tblproperties('skip.header.line.count'='1')

create external table if not exists ext_store_review(
transaction_id string,
store_id string,
review_score int
)
row format delimited fields terminated by ','
location '/tmp/shopping/data/review'
tblproperties('skip.header.line.count'='1')

通过UDF自定义 MD5加密函数

Create MD5 encryption function

这里通过UDF自定义 MD5加密函数 ，对地址、邮箱等信息进行加密。

-- md5 udf自定义加密函数

--add jar /opt/soft/data/md5.jar
--create temporary function md5 as 'com.shopping.services.Encryption'

--select md5('abc')
--drop temporary function encrymd5

Clean and Mask customer_details 创建明细表

create table if not exists customer_details 
as select customer_id,first_name,last_name,md5(email) email,gender,md5(address) address,country,job,credit_type,md5(credit_no) 
from ext_customer_details

对表内容进行检查，为数据清洗做准备

Check ext_transaction_details data
对transaction表的transaction_id进行检查，查看重复的、错误的、以及空值的数量。

这里从表中我们可以看到transaction_id存在100个重复的值。

with 
t1 as (select 'countrow' as status,count(transaction_id) as val from ext_transaction_details),
t2 as (select 'distinct' as status,(count(transaction_id)-count(distinct transaction_id)) as val from ext_transaction_details),
t3 as (select 'nullrow' as status,count(transaction_id) as val from ext_transaction_details where transaction_id is null),
t4 as (select 'errorexp' as status,count(regexp_extract(transaction_id,'^([0-9]{1,4})$',0)) as val from ext_transaction_details)
select * from t1 union all select * from t2 union all select * from t3 union all select * from t4

Clean transaction_details into partition table

create table if not exists transaction_details(
transaction_id string,
customer_id string,
store_id string,
price double,
product string,
buydate string,
buytime string
)
partitioned by (partday string)
row format delimited fields terminated by ','
stored as rcfile

开启动态分区

set hive.exec.dynamic.partition=true
set hive.exec.dynamic.partition.mode=nonstrict

开启动态分区，通过窗口函数对数据进行清洗

Clear data and import data into transaction_details

-- partday 分区 transaction_id 重复 
select if(t.ct=1,transaction_id,concat(t.transaction_id,'_',t.ct-1)) 
transaction_id,customer_id,store_id,price,product,buydate,buytime,date_format(buydate,'yyyy-MM') 
as partday 
from (select *,row_number() over(partition by transaction_id) as ct 
from ext_transaction_details) t

insert into transaction_details partition(partday) 
select if(t.ct=1,transaction_id,concat(t.transaction_id,'_',t.ct-1)) transaction_id,customer_id,store_id,price,product,buydate,buytime,date_format(regexp_replace(buydate,'/','-'),'yyyy-MM') 
as partday from (select *,row_number() over(partition by transaction_id) as ct 
from ext_transaction_details) t 

• row_number() over(partition by transaction_id) 窗口函数 ：从1开始，按照顺序，生成分组内记录的序列，row_number()的值不会存在重复，当排序的值相同时，按照表中记录的顺序进行排列 这里我们对分组的transaction_id
• if(t.ct=1,transaction_id,concat(t.transaction_id,'_',t.ct-1)) 如果满足ct=1,就是transaction_id,否则进行字符串拼接生成新的id

Clean store_review table

create table store_review 
as select transaction_id,store_id,nvl(review_score,ceil(rand()*5)) 
as review_score from ext_store_review

NVL(E1, E2)的功能为：如果E1为NULL，则函数返回E2，否则返回E1本身。

我们可以看到表中的数据存在空值，通过NVL函数对数据进行填充。

show tables

通过清洗后的近源表和明细表如上。

数据分析

Customer分析

• 找出顾客最常用的信用卡

select credit_type,count(credit_type) as peoplenum from customer_details
group by credit_type order by peoplenum desc limit 1

• 找出客户资料中排名前五的职位名称

select job,count(job) as jobnum from customer_details
group by job
order by jobnum desc
limit 5

• 在美国女性最常用的信用卡

select credit_type,count(credit_type) as femalenum from customer_details 
where gender='Female'
group by credit_type
order by femalenum desc
limit 1

• 按性别和国家进行客户统计

select count(*) as customernum,country,gender from customer_details
group by country,gender

Transaction分析

• 计算每月总收入

select partday,sum(price) as countMoney from transaction_details group by partday

• 计算每个季度的总收入
  Create Quarter Macro 定义季度宏，将时间按季度进行划分

create temporary macro 
calQuarter(dt string) 
concat(year(regexp_replace(dt,'/','-')),'年第',ceil(month(regexp_replace(dt,'/','-'))/3),'季度')

select calQuarter(buydate) as quarter,sum(price) as sale 
from transaction_details group by calQuarter(buydate)

• 按年计算总收入

create temporary macro calYear(dt string) year(regexp_replace(dt,'/','-'))

select calYear(buydate) as year,sum(price) as sale from transaction_details group by calYear(buydate)

• 按工作日计算总收入

create temporary macro calWeek(dt string) concat('星期',dayofweek(regexp_replace(dt,'/','-'))-1)

select concat('星期',dayofweek(regexp_replace(buydate,'/','-'))-1) as week,sum(price) as sale 
from transaction_details group by dayofweek(regexp_replace(buydate,'/','-'))

• 按时间段计算总收入(需要清理数据)

select concat(regexp_extract(buytime,'[0-9]{1,2}',0),'时') as time,sum(price) as sale from transaction_details group by regexp_extract(buytime,'[0-9]{1,2}',0)

• 按时间段计算平均消费
  Time macro

create temporary macro calTime(time string) if(split(time,' ')[1]='PM',regexp_extract(time,'[0-9]{1,2}',0)+12,
if(split(time,' ')[1]='AM',regexp_extract(time,'[0-9]{1,2}',0),split(time,':')[0]))

select calTime(buytime) as time,sum(price) as sale from transaction_details group by calTime(buytime) 

--define time bucket 
--early morning: (5:00, 8:00]
--morning: (8:00, 11:00]
--noon: (11:00, 13:00]
--afternoon: (13:00, 18:00]
--evening: (18:00, 22:00]
--night: (22:00, 5:00] --make it as else, since it is not liner increasing
--We also format the time. 1st format time to 19:23 like, then compare, then convert minites to hours
with
t1 as
(select calTime(buytime) as time,sum(price) as sale from transaction_details group by calTime(buytime) order by time),
t2 as
(select if(time>5 and time<=8,'early morning',if(time >8 and time<=11,'moring',if(time>11 and time <13,'noon',
if(time>13 and time <=18,'afternoon',if(time >18 and time <=22,'evening','night'))))) as sumtime,sale 
from t1)
select sumtime,sum(sale) from t2 
group by sumtime

• 按工作日计算平均消费

select concat('星期',dayofweek(regexp_replace(buydate,'/','-'))-1) 
as week,avg(price) as sale from transaction_details 
where dayofweek(regexp_replace(buydate,'/','-'))-1 !=0 and dayofweek(regexp_replace(buydate,'/','-'))-1 !=6
group by dayofweek(regexp_replace(buydate,'/','-'))

• 计算年、月、日的交易总数

select buydate as month,count(*) as salenum from transaction_details group by buydate

• 找出交易量最大的10个客户

select c.customer_id,c.first_name,c.last_name,count(c.customer_id) as custnum from customer_details c
inner join transaction_details t
on c.customer_id=t.customer_id
group by c.customer_id,c.first_name,c.last_name
order by custnum desc
limit 10

• 找出消费最多的前10位顾客

select c.customer_id,c.first_name,c.last_name,sum(price) as sumprice from customer_details c
inner join transaction_details t
on c.customer_id=t.customer_id
group by c.customer_id,c.first_name,c.last_name
order by sumprice desc
limit 10

• 统计该期间交易数量最少的用户

select c.customer_id,c.first_name,c.last_name,count(*) as custnum from customer_details c
inner join transaction_details t
on c.customer_id=t.customer_id
group by c.customer_id,c.first_name,c.last_name
order by custnum asc
limit 1

• 计算每个季度的独立客户总数

select calQuarter(buydate) as quarter,count(distinct customer_id) as uninum
from transaction_details
group by calQuarter(buydate)

• 计算每周的独立客户总数

select calWeek(buydate) as quarter,count(distinct customer_id) as uninum
from transaction_details
group by calWeek(buydate)

• 计算整个活动客户平均花费的最大值

select sum(price)/count(*) as sale
from transaction_details
group by customer_id
order by sale desc
limit 1

• 统计每月花费最多的客户

with 
t1 as
(select customer_id,partday,count(distinct buydate) as visit from transaction_details group by partday,customer_id),
t2 as
(select customer_id,partday,visit,row_number() over(partition by partday order by visit desc) as visitnum from t1)
select * from t2 where visitnum=1 

• 统计每月访问次数最多的客户

with
t1 as
(select customer_id,partday,sum(price) as pay from transaction_details group by partday,customer_id),
t2 as
(select customer_id,partday,pay,row_number() over(partition by partday order by pay desc) as paynum from t1)
select * from t2 where paynum=1

• 按总价找出最受欢迎的5种产品

select product,sum(price) as sale from transaction_details 
group by product
order by sale desc
limit 5

• 根据购买频率找出最畅销的5种产品

select product,count(*) as num from transaction_details 
group by product
order by num desc
limit 5

• 根据客户数量找出最受欢迎的5种产品

select product,count(distinct customer_id) as num from transaction_details
group by product
order by num desc
limit 5

• 验证前5个details

select * from transaction_details where product in ('Goat - Whole Cut')

Store分析

• 按客流量找出最受欢迎的商店

with 
t1 as (select store_id,count(*) as visit from transaction_details 
group by 
store_id order by visit desc limit 1)
select s.store_name,t.visit 
from t1 t 
inner join 
ext_store_details s 
on t.store_id=s.store_id

• 根据顾客消费价格找出最受欢迎的商店

with 
t1 as (select store_id,sum(price) as sale from transaction_details 
group by 
store_id order by sale desc limit 1)
select s.store_name,t.sale 
from t1 t 
inner join 
ext_store_details s 
on t.store_id=s.store_id

• 根据顾客交易情况找出最受欢迎的商店

with
t1 as 
(select store_id,store_name from ext_store_details)
select t.store_id,store_name,count(distinct t.customer_id) as num
from transaction_details t
inner join t1 s
on s.store_id=t.store_id
group by t.store_id,store_name
order by num desc
limit 1

• 根据商店和唯一的顾客id获取最受欢迎的产品

with
t1 as (select store_id,product,count(distinct customer_id) as num from transaction_details
group by store_id,product order by num desc limit 1)
select s.store_name,t.num,t.product 
from t1 t 
inner join 
ext_store_details s 
on t.store_id=s.store_id

• 获取每个商店的员工与顾客比

with
t1 as (select store_id,count(distinct customer_id) as num from transaction_details
group by store_id )
select s.store_name,employee_number/num as vs from t1 t
inner join ext_store_details s 
on t.store_id=s.store_id

• 按年和月计算每家店的收入

select store_id,partday,sum(price) from transaction_details group by store_id,partday

• 按店铺制作总收益饼图

select store_id,sum(price) from transaction_details group by store_id

• 找出每个商店最繁忙的时间段

with
t1 as
(select store_id,count(customer_id) as peoplenum from transaction_details group by store_id,concat(regexp_extract(buytime,'[0-9]{1,2}',0),'时')),
t2 as
(select store_id,peoplenum,row_number() over(partition by store_id order by peoplenum desc) as peo from t1 )
select t.store_id,e.store_name,t.peoplenum from t2 t
inner join ext_store_details e
on e.store_id = t.store_id
where peo =1

• 找出每家店的忠实顾客

with
t1 as
(select customer_id,store_id,count(customer_id) as visit from transaction_details group by store_id,customer_id ),
t2 as
(select customer_id,store_id,visit,row_number() over(partition by store_id order by visit desc) as most from t1)
select r.customer_id,concat(first_name,last_name) as customer_name,r.store_id,store_name,r.visit from t2 r
inner join customer_details c
on c.customer_id=r.customer_id
inner join ext_store_details e
on e.store_id=r.store_id
where most=1

• 根据每位员工的最高收入找出明星商店

with
t1 as
(select store_id,sum(price) as sumprice from transaction_details group by store_id)
select t.store_id,s.store_name,sumprice/employee_number as avgprice  from t1 t
inner join ext_store_details s
on s.store_id=t.store_id
order by avgprice desc

Review分析

• 在ext_store_review中找出存在冲突的交易映射关系

select t.transaction_id,t.store_id from transaction_details t
inner join ext_store_review e
on e.transaction_id=t.transaction_id
where e.store_id!=t.store_id

• 了解客户评价的覆盖率

with
trans as (select store_id,count(transaction_id) as countSale from transaction_details group by store_id),
rev as (select store_id,count(distinct transaction_id) as review from store_review group by store_id)
select s.store_name,(r.review*100/t.countSale) as cover from  trans t 
inner join rev r 
on t.store_id=r.store_id 
inner join ext_store_details s
on t.store_id=s.store_id

• 根据评分了解客户的分布情况

select store_id,review_score,count(review_score) as numview from ext_store_review  where review_score>0 group by review_score,store_id

• 根据交易了解客户的分布情况

select store_id,count(transaction_id) as transactionnum from ext_store_review  group by store_id

• 客户给出的最佳评价是否总是同一家门店

select store_id,customer_id,count(customer_id) as visit from transaction_details t
join ext_store_review e
on e.transaction_id = t.transaction_id
where e.review_score=5
group by t.store_id,t.customer_id