=Paper= {{Paper |id=Vol-3318/short2 |storemode=property |title=Prediction of Mobile Coupon Use: Data Analytics of Influencing Factors |pdfUrl=https://ceur-ws.org/Vol-3318/short2.pdf |volume=Vol-3318 |authors=Shiqui Liu,Dickson K. W. Chiu,Kevin K. W. Ho |dblpUrl=https://dblp.org/rec/conf/cikm/LiuC022 }} ==Prediction of Mobile Coupon Use: Data Analytics of Influencing Factors== https://ceur-ws.org/Vol-3318/short2.pdf

Prediction of Mobile Coupon Use:
Data Analytics of Influencing Factors
Shiqi Liu 1, Dickson K.W. Chiu 1 and Kevin K.W. Ho2
1
University of Hong Kong, Pokfulam Road, Pokfulam, Hong Kong
2
University of Tsukuba, Tokyo, Japan

Abstract
As the Online to Offline (O2O) business model has entered a stage of rapid development,
mobile coupons have attracted the attention of scholars and the industry as a preferential means
to attract customers to pay online. This research uses algorithms and models in machine
learning to explore the consumption records of the mobile coupon data set. Our results show
that the characteristics of merchants and coupons significantly affect the use of mobile
coupons. Based on our findings, a Customer CRM system architecture suitable for merchants
to manage coupons was designed to optimize the business management of coupons and users.

Keywords 1
Mobile coupons prediction; XGBoost; Model; Influencing factors; CRM

1. Introduction

With the latest development in mobile technologies, m-commerce has entered a stage of rapid
development [1]. For example, the e-commerce information, transaction, and technology service
companies in China continue to emerge, with e-commerce transactions reaching 28.4 trillion Yuan in
2018 (equivalent to US$4.5 trillion in 2018), with a year-on-year increase of 17.8%. Online to Office
(O2O) business, as one of the emerging e-commerce business models, entered a rapid development
stage in 2013, beginning to integrate and improve localization and mobile devices [2]. Therefore, the
O2O business model has attracted the attention of researchers and practitioners. Mobile coupons play
an essential role in the O2O business model as a mobile marketing method. It is observed that O2O
merchants use coupons more favorable than offline payments to attract customers to pay online.
Mobile coupons are usually presented in text, images, audio, and video. Customers can get cash
discounts or rebates by showing them when paying. At the same time, they can be used as a carrier of
information activities to drive customers into the store for future purchases [3]. Mobile coupons also
improve customer satisfaction and retention, leading to an increase in the stores’ profit. With the help
of big data technology, these mobile transaction data can be easily analyzed by researchers.
O2O is an emerging e-commerce business model that combines the Internet with offline services,
and the Internet serves as the front desk for transactions. Offline physical stores in this model can attract
customers through the Internet, and customers can search for the services they need. Transactions can
be completed online, quickly forming a scale. The most important feature of this model is: the service
effect can be checked, and every transaction can be tracked and recorded. Thus, this study explores the
influencing factors of mobile coupon use behavior, verifies the influence of coupon discount rate and
reduction on consumers’ willingness to use, and shows that consumers with higher price-sensitive
demand elasticity are willing to pay search costs to reduce the use of mobile coupons.
Under this premise, mobile marketing plays an essential role in facilitating transactions [4]. Mobile
marketing attracts customers to consume by sending service information to customers’ mobile devices.
With the rapid development of network technology and mobile devices, personalized marketing has

Proceedings of THECOG 2022, 10–21, 2022, Atlanta, USA
EMAIL: u3573686@connect.hku.hk (Shiqi Liu); dicksonchiu@ieee.org (Dickson K.W. Chiu); ho.kevin.ge@u.tsukuba.ac.jp (Kevin K.W.
Ho)
ORCID: 0000-0002-7926-9568 (Dickson K.W. Chiu); 0000-0003-1304-0573 (Kevin K.W. Ho)
© 2020 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)
taken shape, and mobile marketing has developed rapidly. Customers search for and understand product
information through various third-party software on mobile phones and tablets, official accounts on
social platforms, text messages, mobile websites, etc., deeply participate in the entire service process,
and finally make a purchase. Mobile coupons are one of the mobile marketing methods under the O2O
business model.

2. Methodology
The data for this research comes from Alibaba Cloud [5] “O2O Coupon Usage”. This data provides
the actual online and offline consumption behavior of a certain platform user between January 1 and
June 30, 2016. For research, we selected a data set of users’ offline consumption and coupon-receiving
behaviors. The information in the data set includes coupon ID, merchant ID, user ID, discount rate, the
distance between the user’s frequent activities and the nearest portal of the merchant, coupon collection
time, and consumption time. There are 1,754,884 records in the original data set, including 8,415
merchants, 539,438 users, and over 1 million coupons.
We used the data set to construct a prediction model for using mobile coupons to predict users’ use
within 15 days after receiving the coupons. We considered three algorithms to conduct our predictive
classification algorithms, i.e., logistic regression, random Forest, and Xgboost [6][7]. In the case of a
single model, Xgboost performs better than the other two methods, and the prediction results are more
accurate (see Table 1).
Table 1
Model score
Model Name Random Forest Logistic Regression Xgboost
Result 0.6932 0.7074 0.8093

3. Influencing factors of the use of mobile coupons
3.1. Analysis of the feature importance score of each module

Figure 1: Histogram of feature importance scores

As Xgboost can more accurately predict the use of coupons, we used it to build a model of whether
mobile coupons are used or not under 52 features. XGboost calculates the number of splits of each
feature in each tree to obtain a feature importance score, indicating the importance of each feature to
model training. Figure 1 shows the result, with the horizontal and vertical axis as the order of the 52
features and the feature score, respectively. Notably, the broken line has a rapid downward trend in the
first sixteen features but declines slowly after the fifteenth feature. This feature is the turning point of
the overall trend of the broken line, and the subsequent feature scores are low, and the effect on the use
of mobile coupons is relatively weak. Therefore, the higher scores, mainly for the first fifteen, have a
more significant impact on mobile coup use.

3.2. Analysis of influencing factors of mobile coupon use

Figure 2: Example figure

Table 2
Feature score ranking
Ranking Name of Feature Type of Feature
1 The total number of historical merchandise sold by Merchant feature
the merchant
2 The merchant’s historical coupon usage rate Merchant feature
3 The total number of coupons issued by the merchant Merchant feature
in the history
4 Discount rate Coupon feature
5 Coupon collection time and feature set deadline time Coupon features
interval
6 Proportion of coupons used in the merchandise sold Merchant feature
by this merchant
7 The average value of the distance between the User-merchant combination
merchant and the user feature
8 Fully reduce the minimum consumption of coupons Coupon feature
9 The month when the coupon is collected Other Features
10 The number of commodities that the merchant has Merchant feature
used coupons to consume
11 The total number of coupons received by users in the User features
month
12 The distance between the user and the merchant User-merchant combination
feature
13 Fully reduce the amount of discount coupons Coupon feature
14 The probability that the user buys goods at the User feature
merchant
15 The number of commodities that the user has used User feature
coupons to consume

After exploring the first fifteen characteristics, it reveals that merchant and coupon characteristics
account for a large proportion of the characteristic score. Table 2 reports the details and ranking and
details of the first fifteen characteristics, and Figure 3 shows the proportion of various characteristics.
The characteristic score accounted for the most significant proportion of merchant characteristics,
accounting for 38%, twice as large as other characteristics with a minor proportion. The coupon
characteristics accounted for the second largest proportion, 26%. User merchant characteristics, user
characteristics, and other characteristics account for relatively small proportions.
The total number of coupons, the proportion of coupons used in the products sold by the merchant,
the average distance between the merchant and the user, and the number of products consumed by the
merchant using coupons. These features are merchant characteristics and indicate the size of the store.
Strength and past sales will significantly affect the use of coupons, indicating that coupons depend on
most regular customers.
Discount rate, coupon collection time and feature set deadline time interval, minimum consumption
amount of full discount coupon, the month of coupon collection time, and full discount coupon discount
amount are features of mobile coupons. The discount rate ranks fourth in the ranking of influencing
factors, indicating that users still pay attention to discounts when using coupons and from the analysis
of consumer psychology. Thus, the greater the discount, the greater the perceived benefits for
consumers and the possibility of using mobile coupons. The factor of coupon collection time and
feature-set deadline time is ranked fifth, indicating that consumers pay attention to the deadline when
receiving mobile coupons. Coupon verification avoids missing the expiration date, so this information
can reflect that a specific store will reach a small climax in sales within a few days of a specific coupon’s
expiration date.
The feature of the total number of coupons received by users in the month belongs to other features
in the analysis. This feature is extracted from the prediction set and ranks thirteenth in the importance
ranking. It shows how many coupons the user received in the month compared to using a certain coupon.
The more coupons users receive, the more time and energy spent on coupon searching and the need to
use coupons to reduce costs.
The distance between the user and the merchant and the probability that the user purchases goods at
the merchant belong to the combined characteristics of the analysis. They are in the twelfth and
fourteenth positions. Generally speaking, the greater the probability of a user buying goods at the
merchant. Therefore, the closer the user and the merchant are, the more trust the user has in the
merchant, thus the higher the possibility of using the merchant’s coupons. The characteristics of the
distance between the user and the merchant indicate that geographic factors will also affect the use of
mobile coupons.
The number of products consumed by the user using coupons and the total number of coupons
received are user characteristics, ranking fifteenth and sixteenth, respectively. The user’s usual coupon
use behaviors also affect the current discounts received. If users use mobile coupons when shopping,
they can save time and energy in coupon search and use and are thus more willing to use received
coupons.

3.3. In-depth exploration of influencing factors of mobile coupon
usage behaviors
After analyzing the factors that affect the use of mobile coupons, we found that these two types of
factors, merchant characteristics and coupon characteristics, affect the behavior of mobile coupons to a
greater extent. Based on the existing data set, we conducted a survival analysis model of merchant and
coupon characteristics. According to the statistical results, mobile coupons with fewer than 30 days of
survival were selected as the research object.
Then, we divided each feature into different groups according to their size. The survival analysis
results differ when the boundary values are different during the grouping process. The median survival
time was used as the observation standard. The group with the most significant change in the median
lifetime was recorded as the grouping result in the table.
Finally, we used Python and SPSS to process survival analysis data. According to the grouping
results in Table 3, each feature was divided into groups with the control group in the survival analysis.
Then, the average survival time of each group’s mobile coupon was obtained, yielding the median
survival time, the significance test results of the adjacent two groups, and the survival curve. After such
repeated steps, the survival analysis of each feature was completed, the numerical experimental results
were recorded in the table, and the survival curve generated by the control group of each feature was
depicted on the same graph.

Table 2
Feature grouping and analysis results
Data Data Group Records Average Median Significance
Group Content Survival
Proportion of coupons used in sold products (Merchant Features)
#1 0-0.03 5,934 9.93 7 < 0.001
#2 0.03-0.06 1,585 7.91 6 < 0.001
#3 0.06-0.16 2,251 7.06 5 < 0.001
#4 0.16-1 941 5.71 3 < 0.001
Use rate of historical coupons (Merchant Features)
#1 0-0.03 1,724 8.92 7 0.004
#2 0.03-0.14 2,246 8.25 6 < 0.001
#3 0.14-0.25 1,890 6.81 5 < 0.001
#4 0.25-1 1,136 5.76 4 < 0.001
The discount rate is different, and the amount of the full discount coupon is the same (Coupon
Feature)
#1 5-1 636 6.98 5 < 0.001
#2 20-1 8,248 8.42 6 0.001
#3 50-10 309 9.93 8 0.001
The discount rate is the same, but the minimum consumption amount for full discount coupons is
different (Coupon Features)
#1 5-1 636 6.98 5 < 0.001
#2 50-10 1,214 9.32 7 < 0.001
#3 100-20 1,000 10.61 9 < 0.001

4. O2O coupons CRM system design
The architecture of our O2O coupon CRM system is classified according to user characteristics and
life cycle management and follows the alert-driven approach to push relevant coupons to potential
customers [8]. Customers are divided into new and regular customers according to their amount,
consumption time, coupons, frequency, and consumption amount. Regular customers are divided into
high-frequency and high-value users, low-frequency and high-value users, silent users, and lost users.
According to different users, businesses can use different strategies.
The O2O coupons use a prediction platform. The current functions are relatively simple, including
four functional modules: user management, coupon management, coupon use prediction, and result
analysis, as shown in Figure 4.
1. User management: This functional module manages users who use this forecasting system,
including adding user registration information, setting account passwords, editing and modifying
user information, and granting user permissions. The system contains two roles. One is the manager,
who can authorize the general user and control the functions that the general staff can access and
use by restricting the menu that the general staff can see. The other is general personnel, who use
customer coupon forecasting and other services within the scope of authority given by the
administrator.
2. Customer management: Edit and enter the customer’s coupon receipt and usage status, and
provide the delete function. These data are the data source for coupon usage prediction.
3. Coupon usage prediction: This functional module predicts the use of coupons, starts to analyze
the data of customers receiving and using coupons, gives some charts for display, improves user
experience, and preprocesses the data in the background, including the processing of missing data,
standardized data, data cleaning, etc. According to the prediction model selected by the user, model
prediction is performed to obtain the prediction result.
4. Result analysis: This module displays the predicted results of customer coupon usage and
compares the real customer coupon usage with the forecast data for users to consult and analyze.

Figure 3: CRM system architecture diagram

5. Conclusion

Starting from user characteristics, merchant characteristics, user merchant combination
characteristics, coupon characteristics, and other characteristics, this research has explored mobile
coupon usage behavior. Based on previous research on mobile coupons, we have constructed a
predictive model and system architecture of mobile coupon use and explored the influencing factors of
mobile coupon use based on the characteristic analysis model.
The research results show that Xgboost has the best effect on predicting results; merchant
characteristics and coupon characteristics significantly affect mobile coupon usage behavior; user
merchant combination characteristics, user characteristics, and other characteristics have an impact on
mobile coupon usage behavior. This research uses supervised machine learning and survival model
analysis methods to compare offline consumption records within half a year. However, there are some
limitations to the research.
1. Due to the limitations of objective conditions, this study’s sample data may be insufficient, and
the sample is not sufficiently representative. The research data shows users’ offline consumption
and coupon receipt behaviors between January 1 and June 30, 2016, but the sales of many other
products are affected by the season. The conclusions obtained from this data sample apply to the
mobile coupon usage behavior in the first and second quarters. In future research, the sample size
can be further expanded to make the sample more representative and scientific.
2. Many factors affect the behavior of using mobile coupons. This research mainly explores the
perspectives of user characteristics, merchant characteristics, coupon characteristics, etc. In contrast,
the existing research on mobile coupon use intention mainly starts from the psychological
characteristics of consumers, such as the perception of usefulness, gender, perceived ease of use,
trust, etc. Future research should include psychological characteristics of consumption to obtain
more comprehensive and practical.
3. The data studied is a data set of users’ offline consumption and coupon receipt behavior,
ignoring online users’ mobile coupon usage behavior. Future research can collect online data to
enrich and triangulate the results.
4. This study combines domestic and foreign mobile coupon research results to explore domestic
consumption data. There is no comparison with foreign data of the same type. When conducting
related research in the future, it is necessary to overcome the limitations of geographic factors and
collect comprehensive data. , To compare the influence of different cultures on the behavior of using
mobile coupons.

6. References
[1] Ni, J., Chiu, D.K.W., Ho, K.K.W. (2022) Exploring Information Search Behavior among Self-
Drive Tourists, Information Discovery and Delivery, in press, doi:10.1108/IDD-05-2020-0054.
[2] Yao, P., Osman, S., Sabri, M. F., & Zainudin, N. (2022). Consumer Behavior in Online-to-Offline
(O2O) Commerce: A Thematic Review. Sustainability, 14(13), 7842.
[3] Liu, F., Liu, S., & Jiang, G. (2022). Consumers’ decision-making process in redeeming and sharing
behaviors toward app-based mobile coupons in social commerce. International Journal of
Information Management, 67, 102550.
[4] Cheung, V.S.Y., Lo, J.C.Y., Chiu, D.K.W., Ho, K.K.W. (2022) Predicting Facebook’s influence
on travel products marketing based on the AIDA model, Information Discovery and Delivery, in
press, doi:10.1108/IDD-10-2021-0117
[5] Kim, Y., Wang, Q., & Roh, T. (2021). Do information and service quality affect perceived privacy
protection, satisfaction, and loyalty? Evidence from a Chinese O2O-based mobile shopping
application. Telematics and Informatics, 56, 101483.
[6] Wang, B., Chiu, D.K.W., Ho, K.K.W (2022) A Comparison of Deep Learning Models in Time
Series Forecasting of Web Traffic Data From Kaggle. In: Sun, Z. (Ed) Handbook of Research on
Foundations and Applications of Intelligent Business Analytics, 301-319, IGI Global.
[7] Gao, W, Lam, K.M., Chiu, D.K.W., Ho, K.K.W (2021) A Big data Analysis of the Factors
Influencing Movie Box Office in China, In: Sun, Z. (Ed) Handbook of Research on Intelligent
Analytics with Multi-Industry Applications, 232-249, IGI Global
[8] Chan, M.M.W., Chiu, D.K.W. (2022) Alert Driven Customer Relationship Management in Online
Travel Agencies: Event-Condition-Actions rules and Key Performance Indicators In A. Naim &S.
Kautish (Ed), Building a Brand Image Through Electronic Customer Relationship Management.
IGI Global, Chapter 12, 250-268.