E-commerce data analysis -- User Behavior Analysis

E-commerce data analysis – User behavior analysis

Data analysis process ：

1. Clear purpose
2. get data
3. Data exploration and preprocessing
4. Analyze the data
5. Come to the conclusion
6. Verification conclusion
7. Result presentation

User behavior refers to the behavior generated by the user on the product .（ land 、 Browse 、 Buy 、 Add to cart ）

User behavior data statistics ：

• Daily active users
• Number of successful registrations
• Number of successful account login
• Number of people entering the product details page
• Number of people added to shopping cart
• Number of successful order submission

Analysis methods

How to conduct user behavior analysis to improve sales ？
Overall analysis of user shopping behavior （ Basic data analysis , Provide data support ）

• PV、UV （ Count by cycle ）
• Average visits （ Count by cycle ）
• Jump loss rate （ Number of click behavior users /UV）
• The date and time period when the user is most active ( Every day 、 Wait a week )

Analysis of commodity purchase （ User preference analysis ）

• Number of purchases （ Count by cycle ）
• Product hits （ Count by cycle ）
• Total analysis of commodity purchase times and clicks （ Carry out summary statistics in the same cycle ）
• Commodity conversion rate （ Purchases and hits ）

User fragrance conversion funnel analysis （ User text path analysis ）

• Click the conversion rate of additional purchase （ Number of buyers / Number of clicks ）
• Click the collection conversion rate （ Number of collectors / Number of clicks ）
• Click purchase conversion （ Number of buyers / Number of clicks ）

methodology （AARR Model）

The data packet , Use Pandas Medium Groupby.（df.groupby(‘str1’,‘str2’).sum()）
AARR Model

Retain users ： Start using the product at a certain time , Users who continue to use the product after a period of time .
Retention ： The number of users who still use the product / Initial total number of users
Repeat purchase rate ： It refers to the proportion of total consumption users who consume twice or more within a certain time window
Repo rate ： Users who consume in a certain time window , Proportion of consumption in the next time window .

E-commerce vulnerability analysis – node Click on 、 purchased 、 Collection 、 Buy

Project operation

Practical operation process

1. Data exploration
2. Data cleaning
3. Data type conversion
4. Build the model
5. Data analysis
6. Interpretation of results

Source code

import time
from datetime import datetime,timedelta,date
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
# Solve the problem of Chinese display 
plt.rcParams['font.sans-serif']=['SimHei']
plt.rcParams['axes.unicode_minus']=False
# Import data 
data = pd.read_csv(r"E:\customer_behavior.csv")
print(data.head())
del data['Unnamed: 0']
print(data.head())
# Analyze the double 11 Activities 
data1 = data[(data.buy_time>='2019-11-05')&(data.buy_time<='2019-11-13')]

print(data1['be_type'].drop_duplicates())
print(data1.isnull().any())# Whether there are missing values 

# Data processing 
data1['day_id']=pd.to_datetime(data1['day_id'],format='%Y-%m-%d')

data1['month']=data1['day_id'].dt.month
data1['buy_time']=data1['day_id'].dt.date
data1['times']=data1['day_id'].dt.time
data1['hours']=data1['day_id'].dt.hour
data1['weekday']=data1['day_id'].dt.dayofweek+1
# View the processed data 
print(data1.head())

#AARR model
# User behavior grouping statistics 
behavior_count = data1.groupby('be_type')['cust_id'].count()

PV=behavior_count['pv']
print("PV=%d"%PV)
UV=len(data['cust_id'].unique())
print("UV=%d"%UV)
print(" Average visits  PV/UV=%d"%(PV/UV))

data_pv=data1.loc[data1['be_type']=='pv',['cust_id']]   
data_fav=data1.loc[data1['be_type']=='fav',['cust_id']]
data_cart=data1.loc[data1['be_type']=='cart',['cust_id']]
data_buy=data1.loc[data1['be_type']=='buy',['cust_id']]
# Set subtraction , Get the number of users with only click behavior 
data_pv_only=set(data_pv['cust_id'])-set(data_fav['cust_id'])-set(data_cart['cust_id'])-set(data_buy['cust_id'])
pv_only=len(data_pv_only)
print(' The tripping rate is ：%.2f%%'%(pv_only/UV*100))

pv_day= data1[data1.be_type=='pv'].groupby('buy_time')['be_type'].count()

uv_day=data1[data1.be_type=='pv'].drop_duplicates(['cust_id','buy_time']).groupby('buy_time')['cust_id'].count()

attr = pv_day.index
v1 = pv_day.values
v2 = uv_day.values

# Set the line width 
plt.figure(figsize=(8, 5))
plt.plot(attr,v1,linewidth=4)
# Set chart title , And label the axis 
plt.title(" Daily hits trend chart ",fontsize=20)
plt.xlabel(' date ',fontsize=10)
plt.ylabel('PV',fontsize=10)
plt.show()

# Line drawing 
fig = plt.figure(figsize=(6,6))      
plt.subplot(2,1,1)  
xlabel=attr
plt.plot(range(len(xlabel)),v1)   
plt.xticks(np.arange(9),'')  
plt.title(' Daily hits trend chart ')
plt.ylabel(' Japan pv')
for a,b in zip(range(len(xlabel)),v1):
    plt.text(a, b, b, ha='center', va='bottom', fontsize=10)   
plt.subplot(2,1,2)
plt.plot(range(len(xlabel)),v2) 
plt.xticks(np.arange(9),('11-05  Tuesday ','11-06  Wednesday ','11-07  Thursday ','11-08  Friday ','11-09  Saturday ','11-10  Sunday ','12-11  Monday ','11-12  Tuesday ','11-13  Wednesday '),rotation=45)
plt.title(' Trend chart of daily number of independent visitors ')
plt.ylabel(' Japan uv')
for a,b in zip(range(len(xlabel)),v2):
    plt.text(a, b, b, ha='center', va='bottom', fontsize=10)
plt.show()

pv_hour=data1.groupby('hours')['cust_id'].count().reset_index().rename(columns={
    ' user ID':'pv'})
uv_hour=data1.groupby('hours')['cust_id'].apply(lambda x:x.drop_duplicates().count()).reset_index().rename(columns={
    ' user ID':' when uv'})

fig,axes=plt.subplots(2,1,sharex=True)
pv_hour.plot(x='hours',y='cust_id',ax=axes[0])
uv_hour.plot(x='hours',y='cust_id',ax=axes[1])
plt.xticks(range(24),np.arange(24))
axes[0].set_title(' Hourly hits trend chart ')
axes[1].set_title(' Trend chart of number of independent visitors per hour ')

def cal_retention(data,n): #n by n Daily retention 
    user=[]
    date=pd.Series(data.buy_time.unique()).sort_values()[:-n] # The time is intercepted before the last day n God 
    retention_rates=[]
    for i in date:
        new_user=set(data[data.buy_time==i].cust_id.unique())-set(user) # Identify new users , In this case, the initial number of users is set to zero 
        user.extend(new_user)  # Add new users to the user group 
        # The first n Day retention 
        user_nday=data[data.buy_time==i+timedelta(n)].cust_id.unique() # The first n Users logged in in days 
        a=0
        for cust_id in user_nday:
            if cust_id in new_user:
                a+=1
        retention_rate=a/len(new_user) # Calculate the day n Daily retention rate 
        retention_rates.append(retention_rate) # Summary n Daily data 
    data_retention=pd.Series(retention_rates,index=date)
    return data_retention

data_retention=cal_retention(data1,3)  # Ask the user 3 Daily retention 

day_buy_user_num = data1[data1.be_type == 'buy'].drop_duplicates(['cust_id', 'buy_time']).groupby('buy_time')['cust_id'].count()
day_active_user_num = data1.drop_duplicates(['cust_id', 'buy_time']).groupby('buy_time')['cust_id'].count()
day_buy_rate = day_buy_user_num / day_active_user_num
attr = day_buy_user_num.index
v1 = day_buy_user_num.values
v2 = day_buy_rate.values

# Set the line width 
plt.figure(figsize=(8, 5))
plt.plot(attr,v1,linewidth=4)
# Set chart title , And label the axis 
plt.title(" Trend chart of daily number of buyers ",fontsize=20)
plt.xlabel(' date ',fontsize=10)
plt.ylabel('PV',fontsize=10)

# Set the line width 
plt.figure(figsize=(8, 5))
plt.plot(attr,v2,linewidth=4)
# Set chart title , And label the axis 
plt.title(" Daily purchase rate trend chart ",fontsize=20)
plt.xlabel(' date ',fontsize=10)
plt.ylabel('PV',fontsize=10)

df_rebuy = data1[data1.be_type == 'buy'].drop_duplicates(['cust_id', 'day_id']).groupby('cust_id')['day_id'].count()
df_rebuy[df_rebuy >= 2].count() / df_rebuy.count()

data_AARR=data1.groupby('be_type')['cust_id'].count()
# Clicks 
pv_value=data_AARR['pv']
# Collection volume 
fav_value=data_AARR['fav']
# Additional purchase 
cart_value=data_AARR['cart']
# Purchase volume 
buy_value=data_AARR['buy']
## Calculate the conversion , There is no order between user collection and additional purchase in actual business , So calculate the conversion rate after the combination of the two 
# Collection Plus purchase conversion rate 
f_c_value=fav_value+cart_value
f_c_ratio=f_c_value/pv_value
print(' The conversion rate of Collection Plus purchase is :%.2f%%'%(f_c_ratio*100))
# Purchase conversion rate 
buy_ratio=buy_value/pv_value
print(' The purchase conversion rate is :%.2f%%'%(buy_ratio*100))

pv_users = data1[data1.be_type == 'pv']['cust_id'].count()
fav_users = data1[data1.be_type == 'fav']['cust_id'].count()
cart_users =data1[data1.be_type == 'cart']['cust_id'].count()
buy_users = data1[data1.be_type == 'buy']['cust_id'].count()
attr = [' Click on ', ' Add to cart ', ' Collection ', ' Buy ']
values = [np.around((pv_users / pv_users * 100), 2),
         np.around((cart_users / pv_users * 100), 2),
         np.around((fav_users / pv_users * 100), 2),
         np.around((buy_users / pv_users * 100), 2)]

from pyecharts.charts import Funnel
funnel1 = Funnel(" Overall transformation funnel ",width=800, height=400, title_pos='center')
funnel1.add(name=' link ',  #  Specify the legend name 
            attr=attr,  #  Specify the property name 
            value=values,  #  Specify the value corresponding to the attribute 
            is_label_show=True,  #  Confirm that the label is displayed 
            label_formatter='{c}'+'%',  #  Specify how labels are displayed 
            legend_top='bottom',    #  Specify the legend location , To avoid covering, choose the lower right display 
            # pyecharts The documentation of the package indicates , When label_formatter='{d}' when , Labels are displayed as a percentage .
            #  But when I do this , It is found that the displayed percentage does not correspond to the original data , I have to use the display form above 

            label_pos='outside',  #  Specify the location of the label ,inside,outside
            legend_orient='vertical',  #  Specify the direction of legend display 
            legend_pos='right')  #  Specify the location of the legend 

funnel1.render_notebook()

product_buy=data1.loc[data1['be_type']=='buy',['cust_id','group_id']]
product_buy_count=product_buy.groupby('group_id')['cust_id'].count().rename(' Number of sales ')

product_buy_count=pd.DataFrame(product_buy_count) 
product_buy_count=product_buy_count.sort_values(by=' Number of sales ',axis=0,ascending = False)

product_buy_count=product_buy_count.iloc[:10,:]
product_buy_count

product_pv=data1.loc[data1['be_type']=='pv',['cust_id','group_id']]
product_pv_count=product_pv.groupby('group_id')['cust_id'].count().rename(' Number of hits ')
product_pv_count=pd.DataFrame(product_pv_count)   
product_pv_count=product_pv_count.sort_values(by=' Number of hits ',axis=0,ascending = False)
product_pv_count=product_pv_count.iloc[:10,:]
product_pv_count

item_behavior=data.groupby(['group_id','be_type'])['cust_id'].count().unstack(1).rename(columns={
    'pv':' Clicks ','fav':' Collection volume ','cart':' Additional purchase ','buy':' Purchase volume '}).fillna(0)
item_behavior.head()
item_behavior[' Conversion rate ']=item_behavior[' Purchase volume ']/item_behavior[' Clicks ']
item_behavior.head()

top=pd.concat([product_buy_count,product_pv_count],axis=1,sort=False)
top

Analysis report ： From the analysis of the above results .
for example ：
1、 The time period with the least hits every morning , The case with the least user visits . Peak user activity , What should be done for this peak period .
2、 The situation and analysis of user repurchase rate .
3、 Order analysis
4、 Funnel analysis