DATA PAPER - IA de Investigación especializada en el MIT

PAPEL DE DATOS

Constructing a Scene-Based Knowledge System
for E-Commerce Industries:
Business Analysis and Challenges

Min Fu1,2, Qiang Chen1, Wei Lin1, Pei Wang1 & Wei Zhang1

1Alibaba Group, Yu Hang District, Hangzhou 311121, Porcelana

2School of Computing, Macquarie University, North Ryde NSW, Sídney 2109, Australia

Palabras clave: E-commerce business; Scene-based; Knowledge system; Product categorization; Business goals;

Business challenges

Citación: METRO. Fu, q. Chen, W.. lin, PAG. Wang, & W.. zhang. Constructing a scene-based knowledge system for E-commerce industries:

Business analysis and challenges. Data Intelligence 1(2019). doi: 10.1162/dint_a_00012

Recibió: December 31, 2018; Revised: Enero 28, 2019; Aceptado: Febrero 2, 2019

ABSTRACTO

Online marketers make efforts to sell more products to their customers to increase turnover. One way to
sell more products is to ensure that products belonging to the same scene are offered together. Tal como, es
beneficial to categorize products into different groups and tag them to particular, defined scenes. El
mechanism of building relationships between products and scenes is called “scene-based knowledge system
construction”. The construction of a scene-based knowledge system is usually based on business operators’
expert knowledge and past experiences, which often causes product categorization to be inaccurate. Por eso,
in this paper, we discuss the importance and necessity of constructing a scene-based knowledge system,
analyze it from a business perspective, and discuss its challenges.

1. INTRODUCCIÓN

In global e-commerce, businesses compete for markets, products, customers, marketplaces, sellers, y
even operational technologies to increase their profits and overall turnover [1]. Every e-commerce business,
especially the large-scale ones, such as Alibaba [2], expends great efforts to prompt consumers to buy more
products from online shopping sites. One of the strategies they use is to ensure that multiple products

† Corresponding author: Min Fu (Correo electrónico: hanhao.fm@alibaba-inc.com; ORCID: 0000-0002-1557-0810).

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
d
norte

t
/

a
r
t
i
C
mi
–
pag
d

F
/

1
3
2
2
4
6
8
3
7
7
4
d
norte
_
a
_
0
0
0
1
2
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Constructing a Scene-Based Knowledge System for E-Commerce Industries: Business Analysis
and Challenges

belonging to the same particular scenario, or scene, are displayed to the customers so that they may
purchase more products. This is also called “item-binding sales [3]". In this mechanism, when a customer
searches for a certain product, other products that share the same scene with that product are shown to
the customer, which may make the customer select and purchase more items. To make this mechanism
successful, retailers need to determine scenes for products; en otras palabras, we need to categorize products
into different groups, in which each group is tagged by a particular scene. en este documento, we call this
“constructing a scene-based knowledge system”.

Scene-based knowledge system construction is important for three reasons: 1) it provides detailed
information on the relationships between products and scenes; 2) it can be used to make recommendations
to buyers or customers when listing products for them, so that they can avoid a new product search; 3) él
helps to bind relevant items together for online shop owners to display products, so that they can sell more
products and hence, increase their profits.

Existing ways of constructing a scene-based knowledge system usually rely on expert knowledge and
past experience of business operators in grouping products into different functional categories [4,5,6].
These methods do not yield a consistent or accurate scene-based knowledge system because they are not
generalizable. Además, relationships between products and scenes are highly dependent on the specific
business scopes and requirements formulated by e-commerce businesses; hence, it is difficult to categorize
products into groups by particular scenes. Como consecuencia, we need more fine-grained methods for constructing
a scene-based knowledge system for e-commerce businesses. In order to propose better methods, nosotros
should first understand the business reasons for constructing a scene-based knowledge system for e-commerce
negocios. Después, we need to clarify its business requirements, business goals, and business value,
followed by an examination of its challenges [7]. Además, we should also propose potential solutions
to address these challenges. en este documento, tal como, we make a business analysis of scene-based knowledge
system construction by discussing its business reasons, requirements, objetivos, and value, and we list its
challenges and attempt to address these challenges by proposing potential solutions.

The research contributions of this paper are as follows: 1) formulating the official definition of a scene-
based knowledge system for e-commerce businesses, which can be used by any e-commerce corporation
in the world; 2) making a detailed business analysis for the construction of a scene-based knowledge
sistema, by analyzing its business reasons, business requirements, business goals and business value; 3)
providing a thorough list of challenges faced in scene-based knowledge system creation for e-commerce
industries, followed by explanations for each challenge and analysis; y 4) proposing a series of potential
solutions that can be applied to addressing these challenges, which can be widely used by e-commerce
businesses worldwide.

The remainder of this paper is organized as follows: Sección 2 discusses the background. Sección 3
discusses related works. Sección 4 presents a motivating example. Sección 5 conducts a business analysis
for scene-based knowledge system construction. Sección 6 lists the challenges with scene-based knowledge
system construction and Section 7 provides potential solutions to the challenges. Sección 8 provides a
discussion, and finally, Sección 9 provides a conclusion and delineates future work.

225

Data Intelligence

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
d
norte

t
/

a
r
t
i
C
mi
–
pag
d

F
/

1
3
2
2
4
6
8
3
7
7
4
d
norte
_
a
_
0
0
0
1
2
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Constructing a Scene-Based Knowledge System for E-Commerce Industries: Business Analysis
and Challenges

2. BACKGROUND

This section consists of two aspects: 1) the definition of a scene-based knowledge system in e-commerce
businesses and 2) the current usage of the scene-based knowledge system in e-commerce industries
worldwide.

2.1 Definition of a Scene-Based Knowledge System

In order to understand and define the term “scene-based knowledge system”, we should understand the
core words: “scene” and “knowledge system”. A scene refers to a scenario where a particular customer
makes a certain purchase for a certain purpose, and a knowledge system refers to the type of knowledge
that is represented by certain relationship information between two entities, such as the relationship between
products and scenes. Por ejemplo, a pair of basketball training shoes is used in the scene of playing
basketball.

Primero, we need to formally define the word “scene”. Específicamente, the scenes in this study are shopping
escenas. While other research works [8,9,10] define shopping scenes as shopping scenarios where people
purchase products in order to meet shopping requirements, such as living needs or educational needs, nosotros
define scenes using the 4-W principle. The 4-W principle consists of four aspects: 1) cuando, 2) OMS, 3)
dónde, y 4) qué. This means that a certain person (OMS) does something (qué) in a certain location
(dónde) at a certain time (cuando). It is not necessary that all four Ws are present in a scene; sin embargo, el
“what” must be present in it. The other three Ws (cuando, who and where) can be combined with what,
either separately or jointly. Por ejemplo, a scene can be “who does what”, “when it is done”, “where it is
done”, or “who does what where and when”. A shopping scene is equivalent to the shopping purpose
because the scene can be reflected in the reason why a customer wants to buy a particular product. Para
ejemplo, the purchase of a pair of basketball training shoes belongs to the scene of playing basketball and
it is used for playing basketball. Por eso, according to our definition, a scene is formulated as shown:

Scene = Tuple(Qué) ∪ {Sw | Sw ∈ Tuple(Cuando, Who, Where)}.

(1)

Próximo, we should formally define the term “knowledge system”. According to the existing works in
knowledge graphing, a knowledge system refers to a certain type of knowledge infrastructure that represents
complete relationships among multiple knowledge nodes [7, 11, 12]. In the context of e-commerce
negocios, we define a knowledge system as the information that depicts relationships between two
knowledge entities, p.ej., products and scenes. We subsequently use this knowledge system to describe the
knowledge architecture within the e-commerce industry.

Finalmente, we need to provide the definition of the complete term “scene-based knowledge system”. Este
term is defined as the data-oriented knowledge architecture that contains all of the mapping relationships
between the two entities of products and scenes. Usually, a knowledge system is regarded as a paradigm
that generally depicts the entities and relationships between entities in a high abstraction level. Since this
scene-based knowledge system is for the purpose of serving all business services and activities within
e-commerce industries, such as Alibaba Group, we use the term “knowledge system” to imply the deep

Data Intelligence

226

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
d
norte

t
/

a
r
t
i
C
mi
–
pag
d

F
/

1
3
2
2
4
6
8
3
7
7
4
d
norte
_
a
_
0
0
0
1
2
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Constructing a Scene-Based Knowledge System for E-Commerce Industries: Business Analysis
and Challenges

functionalities and insights of the use of our product–scene relationships. We denote the scene-based
knowledge system as SBKS, and it can be defined as:

SBKS = {Tuple(Product, Scene) | Product is used in Scene & Scene includes Product}.

(2)

2.2 Current Usage of Scene-Based Knowledge System in E-Commerce Businesses

Several global e-commerce corporations based in China, such as Alibaba Group and JD.Com, tener
constructed scene-based knowledge systems [2, 13]. They apply scene-based knowledge systems to many
areas, including shopping guidance, product recommendations and item-binding sales [2, 13]. By using
this strategy, profits have risen by a substantial amount, overall turnover has greatly increased, and customer
satisfaction has improved. The “Tiny Eastern Courtyard” product developed by JD.Com [13] can be used
as an example to illustrate. This product was developed by JD.Com so as to increase the number of user
visits, user page clicks and the user purchase rate of its online shopping application (app). It contains more
than 1,300 shopping scenes, of which 200 are formally used by the front side of the app [13]. JD used a
combination of intelligent algorithms and business operators’ operational strategies to review the scenes
that should be used and manually built the relationships between each product and scene. Entonces, it applied
the constructed scene-based knowledge system on the functionalities of product recommendations and
shopping guidance. By doing so, its user visit number increased substantially, and the purchase rate of its
products also increased. This product has, por lo tanto, gained a strong positive reputation within JD.Com
Group [13].

Alibaba’s shops are another example that can be used to clarify the current usage of a scene-based
knowledge system; multiple products with various functionalities and purposes are displayed on the main
page and the details pages of each shop. When a customer browses the products listed in an online shop,
it can be difficult to find all products that belong to the same scene, requiring extra time in looking for
other products belonging to the same scene. In order to solve this problem, Alibaba Group developed two
applications by constructing a scene-based knowledge system: an app called “Have Good Things” and
another called “Must-Buy List” [2]. The former product is used for making recommendations to customers
looking for products belonging to the same shopping scenes, and the latter is used to provide insightful
shopping guidance to customers who are not very certain about what products they should buy [2]. El
use of these two apps have led to an increase in the number of user visits, the number of page clicks, y
the purchase rate of Alibaba Group’s online shops [14].

3. RELAT ED WORKS

Since we intend to employ several machine learning techniques to address some of the challenges of
constructing a scene-based knowledge system for e-commerce businesses, we will first discuss some works
related to applying machine learning methods to solving financial and commercial problems. These works
include using machine learning techniques to determine share market prices, to forecast financial time
series, and to predict the price of crude oil [15]. These works are related to our research in terms of using
machine learning to determine a product’s scenes and analyze information about the product’s properties
and the scene’s characteristics and mining the relationship between them.

227

Data Intelligence

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
d
norte

t
/

a
r
t
i
C
mi
–
pag
d

F
/

1
3
2
2
4
6
8
3
7
7
4
d
norte
_
a
_
0
0
0
1
2
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Constructing a Scene-Based Knowledge System for E-Commerce Industries: Business Analysis
and Challenges

3.1 Share Market Price Prediction with Artificial Neural Networks

En 2011, researchers from Shahjalal University of Science and Technology (SUST) proposed a mechanism
of predicting the stock market price based on artificial neural networks [15, 16]. This mechanism addressed
a significant challenge which is related to the insignificant relationships between the share market chaos
system’s variables and the share price [15, 16]. They applied the traditional multiple layer perceptron (MLP)
model on the neural network, and relied on the back propagation (BP) mechanism to compute and update
the weight values of the intermediate inputs and outputs [15, 16]. The main drawbacks with this mechanism
are as follows: 1) the neural network only used two hidden layers and there was no explanation for it, y
they did not make any comparison between the performance of using two hidden layers and using more
hidden layers; 2) the share market’s variables which were used in the neural network only considered the
information about the financial situation of the company, without considering other important features,
such as customer size or launch time [15, 16]. Además, the proposed method was evaluated with only
one company and tested against only two sets of input data, and the validity of the method was not fully
proved [15, 16].

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
d
norte

t
/

a
r
t
i
C
mi
–
pag
d

F
/

1
3
2
2
4
6
8
3
7
7
4
d
norte
_
a
_
0
0
0
1
2
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

3.2 Financial Time Series Forecasting with ANNs

Artificial neural networks (ANNs) were identified as the most powerful and useful machine learning
technique for the prediction of financial time series in the stock market by researchers from the Bond
Universidad, Australia [15, 17]. While the other methods used for such a prediction were majorly based on
evolutionary and optimization-based technologies, people would prefer to employ and intensify established
ANN models with new training algorithms or would like to combine ANNs with emerging technologies
into hybrid systems [16, 18]. Sin embargo, it is still unclear how the real-world constraints can impact the
accuracy of financial time series forecasting and stock index prediction. Además, we should also
investigate whether investors’ risk-return tradeoff can be improved or not [15, 17].

3.3 Crude Oil Price Prediction with ANN-Q

An artificial neural network quantitative (ANN-Q) model-based approach for predicting the price of
crude oil was proposed by researchers from the University of Manchester in 2010 [15, 18]. The framework
proposed in this research addressed a significant challenge with crude oil price prediction: the price of
crude oil is linked with a large set of factors and the changes of these factors may impact the price of crude
oil to a large extent. Crude oil is a major product that has a high level of volatility in the world [15, 18].
By leveraging and analyzing a total of 22 key factors which can impact the price of crude oil, investigadores
proposed an ANN-Q model [15, 18]. For the purpose of developing a better model, the data related to
these key factors were further divided into three categories: large impact, medium impact and small impact
[15, 18]. The Backpropogation Neural Network (BPNN) was utilized for input variables training [15, 18].
In the simulation based experiments, the acceptable accuracy of the proposed method was monitored
progressively, and the results show that the accuracy could be improved by better tuning the parameters of
el modelo [15, 18].

Data Intelligence

228

Constructing a Scene-Based Knowledge System for E-Commerce Industries: Business Analysis
and Challenges

4. A MOTIVATING EXAMPLE

En esta sección, we present a motivating example to explain why constructing a scene-based knowledge
system for e-commerce businesses is beneficial. With a developed scene-based knowledge system, a set of
business requirements and operational strategic goals can be fulfilled by applying the system to an
e-commerce business solution, such as product recommendations and item-binding sales. The motivating
example is illustrated in Figure 1.

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
d
norte

t
/

a
r
t
i
C
mi
–
pag
d

F
/

1
3
2
2
4
6
8
3
7
7
4
d
norte
_
a
_
0
0
0
1
2
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Cifra 1. A motivating example.

A customer is attached to several life scenes, and one of them could be the birth of a baby. The scene
of the birth of a baby consists of several subsidiary scenes, and hence, the customer needs to make a scene
recognition for this main scene. After analyzing the subsidiary scenes of the main scene, the customer
observes that it can be divided into several child scenes: selecting a hospital, getting a pregnancy scan,
prenatal education, making preparations for the birth, post-delivery care, buying newborn insurance, y
ensuring that newborn vaccinations are given to the baby. Entonces, based on these sub-scenes, the customer
can generate many main shopping requirements that break down into several child-related requirements.
This is performed with the assistance of online applications, such as Baidu, Zhihu and WeChat; vertical
sites; e-commerce sites; and the Taobao App. Después, based on these sub-requirements, the customer
determines that they need to purchase more than 70 baby products, including milk powder, diapers and
milk bottles. Próximo, the customer needs to select proper shopping channels from the candidates list, semejante
as the Taobao App, JD.Com, NetEase, and Baobao Tree. Entonces, the customer makes shopping choices by
analyzing the deterministic features of the products, considering product prices, studying the comments

229

Data Intelligence

Constructing a Scene-Based Knowledge System for E-Commerce Industries: Business Analysis
and Challenges

and feedback of the products, etcétera. Finalmente, the customer decides what to buy and places orders for
those products. In the above story, we can see that many products belong to the same scene and the same
subsidiary scene. If we combine the products belonging to the same scene/sub-scene together and put them
into a category tagged by a certain scene, then the customer will find it easier and more convenient to
place orders for products that belong to the same scene. As such, the construction of a scene-based
knowledge system is a necessary element in e-commerce sales for the arrival of a new baby.

5. BUSINESS ANALYSIS FOR SCENE-BASED KNOWLEDGE SYSTEM CONSTRUCTION

This section contains a detailed business analysis for the construction of a scene-based knowledge system.
Específicamente, we provide the business reasons supporting the development of a scene-based knowledge
sistema, formulate its business requirements, illustrate its business goals and discuss its business value.

5.1 Business Reasons for Constructing a Scene-Based Knowledge System

Based on our past experience as well as the expertise of business professionals from other e-commerce
organizaciones, we present eight reasons for a scene-based knowledge system to be constructed. These eight
reasons are: 1) the information about the categories of the shopping scenes helps e-commerce businesses
better understand the complete scenes (p.ej., the shopping scene for baby care includes purchasing infant
alimento, planning infant wellness, and purchasing infant clothing, among other shopping tasks); 2) if features
of online products are analyzed in a scene-oriented manner, the business stakeholders will have better
ideas about which scenes online products belong to (p.ej., basketball shoes’ features should include weight,
materiales, durability and color); 3) the online shopping systems need to know the manner in which online
products should be put together and how these categorized products can be sold to their customers in just
one round (p.ej., the basketball shoes and the basketball shorts should be bound together for sale); 4) el
number of user visits and page clicks as well as the purchase rate of the online shops are highly related
to the scene-based products’ categorization mechanisms (p.ej., after binding the basketball shoes and the
basketball shorts, the overall click rate on these two items and the sales of these two items increased);
5) for customers, the quality of their shopping experience should be improved; this is closely related to
whether related products are categorized according to their scenes and purposes (p.ej., if the basketball
shoes are bound with basketball shorts, customers regard the site as a convenient way to shop for basketball
gear); 6) the total profits and overall turnover of an e-commerce businesses can be increased by using
estrategias, such as product recommendations and item-binding sales that rely on the construction of a
scene-based knowledge system (p.ej., selling only the shoes brings less profit than selling both the basketball
shoes and basketball shorts); 7) there is a lack of standardization in the relationships between online
products and shopping scenes among e-commerce businesses worldwide, and hence, these businesses do
not have a unified mechanism to categorize their products based on particular scenes; 8) the current
definition of the scene-based knowledge system is not formalized and the usage of scene-based knowledge
systems is not standardized among the various e-commerce businesses in the world.

Data Intelligence

230

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
d
norte

t
/

a
r
t
i
C
mi
–
pag
d

F
/

1
3
2
2
4
6
8
3
7
7
4
d
norte
_
a
_
0
0
0
1
2
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Constructing a Scene-Based Knowledge System for E-Commerce Industries: Business Analysis
and Challenges

5.2 Business Requirements for Constructing a Scene-Based Knowledge System

We also formulate the business requirements for constructing a scene-based knowledge system for
e-commerce businesses. The following are the six requirements: 1) the scenes and shopping purposes of
all customers that shop on online shopping sites should be enumerated daily in a complete manner to
ensure we observe shopping scenes that are important for making e-commerce decisions; 2) the definition
of scenes and the relationships among various scenes must be determined in a clear and accurate manner;
3) the mapping relationships between each online product and each scene must be accurately and precisely
determined and those relationships must be easy to understand; 4) the multitude of products that can belong
to an identical scene must be properly defined; 5) if a product can belong to multiple scenes, then all
scenes must be determined; y 6) the constructed scene-based knowledge system must be stored in a
proper form in a reasonable location to be easily accessed.

5.3 Business Goals for Constructing a Scene-Based Knowledge System

The business goals of constructing a scene-based knowledge system are as follows. Primero, the system can
be used by all e-commerce businesses in a generalizable way; different e-commerce businesses are able
to treat it as a standardized tool to make business decisions. Segundo, it can be used by all worldwide
e-commerce businesses to develop sophisticated functionalities of online shopping applications. Tercero,
it will greatly increase user visits and page clicks on online shopping apps through improved product
recommendations and item-binding sales; and finally, it will substantially increase purchase conversion
tarifas, overall profits and total turnover.

5.4 Business Value of Constructing a Scene-Based Knowledge System

The business value of constructing a scene-based knowledge system is three-fold: 1) it provides a standard
for organizing and managing products, scenes and their relationships for all e-commerce businesses
worldwide; 2) it helps to achieve business goals, such as total user visits, user page click rates, user purchase
tarifas, business profits and turnover; y 3) it sets a good example for e-commerce businesses and companies
in other domains for building and using a knowledge system that is closely related to business strategy.

6. CHALLENGES

In this section we list eight constraints in this project of constructing a scene-based knowledge system

for e-commerce as follows.

1). The determining of scenes is not a straightforward process because there is no standardization of
what “a scene” really entails. There is no previous work that can be referenced to help us understand
the internal properties of scenes. Por lo tanto, we have to determine the definitions based on the
understanding of our own e-commerce corporation’s contextual information and business strategies.

231

Data Intelligence

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
d
norte

t
/

a
r
t
i
C
mi
–
pag
d

F
/

1
3
2
2
4
6
8
3
7
7
4
d
norte
_
a
_
0
0
0
1
2
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Constructing a Scene-Based Knowledge System for E-Commerce Industries: Business Analysis
and Challenges

2). There is no existing work on the method to systematically define and formulate a scene-based
knowledge system; hence, we lack references to describe and formulate the system. As such, nosotros
must rely on our own context and business scope to systematically formulate the definition and use
this definition to construct the relationships between products and scenes.

3). The core aspect of a scene is “a thing”, but the definition of “a thing” is quite unclear and different
people have different opinions about what a thing is. Además, the relationship between a thing
and other aspects of a scene is not always straightforward.

4). It is difficult to describe or model an individual customer’s shopping requirements. Sometimes a
customer may not know the actual reasons why they wish to buy something. Shopping requirements
can be quite complex, which makes it complicated to model shopping requirements.

5). The definition of scenes is based on the understanding of operational rules; sin embargo, operational
rules are difficult to define because different business operators have different standards on managing
rules and rules often come in various forms. Por eso, it is challenging to define them in a generic
way.

6). It is also exacting to determine the relationship between each product and each scene because it is
based on a full and global understanding of all products and their features and all scenes and their
propiedades. This takes a relatively long time and substantial human efforts; it is difficult to automate
this process.

7). It is difficult to verify the accuracy of the determined relationships between products and scenes as
it requires manpower and considerable human efforts. It largely depends on the business scope and
requirements of the e-commerce industry. Even if we use automation mechanisms to speed up this
proceso, it is not guaranteed that the determined relationships will be accurate enough.

8). The mechanism used to construct a scene-based knowledge system must be generalizable to all
e-commerce businesses in the world. This is demanding because different e-commerce businesses
have different characteristics, products and product categories.

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
d
norte

t
/

a
r
t
i
C
mi
–
pag
d

F
/

1
3
2
2
4
6
8
3
7
7
4
d
norte
_
a
_
0
0
0
1
2
pag
d

7. POTENTIAL SOLUTIONS

In order to address the abovementioned challenges, we propose potential solutions to solve each of them.

These solutions are described as below:

1). To solve the challenge related to the definition of scenes, we propose to make a business study on
all the purposes our customers have in buying certain products. We can model these drives and
attempt to mine the shopping scene information from the purposes. More specifically, we should
determine all aspects of a scene and clarify which aspects are the core aspects and which are
secondary.

2). To solve the challenge related to the knowledge system definition, we propose to review the business
requirements and operational requirements of our own e-commerce corporation to understand the
type of knowledge system our firm requires to determine all features and aspects that are needed in
formulating the definition of a knowledge system.

b
y
gramo
tu
mi
s
t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Data Intelligence

232

Constructing a Scene-Based Knowledge System for E-Commerce Industries: Business Analysis
and Challenges

3). To solve the challenge of the definition of “things”, we propose to summarize all objective entities
within the purposes of online shopping tasks conducted by all types of customers and try to determine
the patterns that model each entity. Como resultado, the “things” will be the models in the form of proper
patrones.

4). To solve the challenge of modeling shopping requirements, we propose to first gather historical
shopping requirements of all existing customers, either manually or in an automated manner, y
subsequently use data analytics methods to perform data mining and statistical learning on these
data to discover the correct way to model user shopping requirements. This model should be stored
in such a manner that it can be reused by external parties.

5). To solve the challenge of defining the operational rules, we propose to determine a formal way of
expressing an operational rule. To express an operational rule properly and correctly, we should be
certain about what factors are included in the rule and how to organize these factors to form a
scientific and model-oriented paradigm to record the operational rules.

6). To solve the challenge related to the determination of the product–scene relationships, we propose
to use a combination of two mechanisms: 1) using machine learning to generate and classify scenes
y 2) having people check the conformance of the operational rules. Primero, we generate a scene for
each product; segundo, we determine whether the generated scene is proper for the product according
to the conformance checker.

7). To solve the challenge related to the accuracy of product–scene relationships, we propose to apply
cross validation on the determined output of the relationships between products and scenes.
Específicamente, we can use 10-fold cross validation, where we divide the data set into ten sections, usar
nine of the ten sections for data training, and use the remaining section for data prediction to
calculate the prediction accuracy. Performing this procedure ten times until each of the ten sections
has been evaluated, we then compute the average amount of the ten accuracy values and verify
whether this average value is large enough.

8). To solve the challenge of generalizing the system, we propose to first summarize all products and
scenes for each of the worldwide e-commerce businesses and subsequently build the relationships
between products and scenes for each e-commerce business, using the data for each e-commerce
business for a small scene-based knowledge system. Próximo, we will unite all knowledge systems
together to form a larger, centralized scene-based knowledge system. Different e-commerce
businesses will be able to use the same centralized knowledge system to develop their own strategic
solutions for their applications so as to fulfill their respective business goals.

8. DISCUSIÓN

En realidad, the construction of a scene-based knowledge system is only one of the many strategies that
can be used to fulfill the business goals and requirements of e-commerce businesses in the world. Otro
strategies may include marketing solutions, such as intelligent red packets [9], intelligent coupons, intelligent
price determinations [6] or price discounting mechanisms. We argue that a scene-based knowledge system
is more viable than other strategies because we firmly believe that the purposes behind customers’ purchases
of certain products are fundamental to successful e-purchasing.

233

Data Intelligence

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
d
norte

t
/

a
r
t
i
C
mi
–
pag
d

F
/

1
3
2
2
4
6
8
3
7
7
4
d
norte
_
a
_
0
0
0
1
2
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Constructing a Scene-Based Knowledge System for E-Commerce Industries: Business Analysis
and Challenges

While a scene-based knowledge system can be used to realize business goals, such as increasing the
number of user visits, page visits, purchase rates, and overall profits and turnover, it can also offer other
benefits to e-commerce businesses, namely strengthening their reputations and helping with business
expansions. De este modo, it is vital and significant for global e-commerce businesses to set up resources to construct
scene-based knowledge systems.

9. CONCLUSION AND FUTURE WORK

Every online shop owner is dedicated to selling more products to customers to increase the total turnover.
One method of achieving this is to ensure that products belonging to the same scene are grouped together.
As such, firms must categorize products into different groups that are tagged by particular scenes. Nosotros
call the determination of the relationships between products and scenes “scene-based knowledge system
construction”. The existing methods of constructing scene-based knowledge systems are typically based
on the business operators’ expert knowledge and past experience, which results in inaccurate product
categorization. Por eso, in this paper we make a detailed business analysis of scene-based knowledge system
construction, discuss its challenges, and propose potential solutions to these challenges.

Our future work includes the following tasks. Primero, we intend to propose a formal method to construct
a scene-based knowledge system, implementing and evaluating it using real-world scenarios. The method
will be a combination of potential solutions to challenges delineated in Section 7. It must be fine-grained
and generalizable for different types of business requirements and cater to multiple e-commerce industries
in the world. Segundo, we need to determine other business requirements that the scene-based knowledge
system should support and check whether our methodology is able to fulfill those requirements.

CONTRIBUCIONES DE AUTOR

This work was a collaborative effort among all authors. METRO. Fu (hanhao.fm@alibaba-inc.com, correspondiente
author) is the main writer for the research project. q. Chen (lapu.cq@alibaba-inc.com) W.. lin (weijiang.
lw@alibaba-inc.com), PAG. Wang (wp146049@alibaba-inc.com), and W. zhang (lantu.zw@alibaba-inc.com)
provided insights and information in the business analysis section the challenges section, y el
methodology section of the paper. All authors made meaningful and valuable contributions in re vising and
corregir el manuscrito resultante.

REFERENCIAS

2.
3.

S. sharma. Internet marketing: The backbone of e-commerce. International Journal of Emerging Research in
Management & Tecnología 6(4) (2016), 38–51. doi: 10.4018/jebr.2010100104.
Alibaba Group’s official website. Disponible en: https://www.alibaba.com/.
mi. Turban, D. Rey, J.K. Sotavento, T.P. Liang, & D.C. Turban. Electronic commerce: A managerial and social net-
works perspective. Berlina: Saltador, 2015. isbn: 978-3319100906.

Data Intelligence

234

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
d
norte

t
/

a
r
t
i
C
mi
–
pag
d

F
/

1
3
2
2
4
6
8
3
7
7
4
d
norte
_
a
_
0
0
0
1
2
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Constructing a Scene-Based Knowledge System for E-Commerce Industries: Business Analysis
and Challenges

R. Mansell. Constructing the knowledge base for knowledge–driven development. Journal of Knowledge
Management 6(4) (2002), 317–329. doi: 10.1108/13673270210440839.
A. Gomez-Perez, oh. Corcho, & METRO. Fernandez-Lopez. Ontological engineering: With examples from the areas
of knowledge management, e-commerce and the semantic web. 1st edition. Londres: Saltador, 2004. ISBN:
1852338407.
PAG. cocinero, & l. Leydesdorff. Regional development in the knowledge-based economy: The construction of
advantage. Journal of Technology Transfer 31(1) (2006), 5-15. doi: 10.1007/s10961-005-5009-3.
A. Tsalgatidou, & mi. Pitoura. Business models and transactions in mobile electronic commerce: Requirements
and properties. Journal of Computer Networks 37(2) (2001), 221–236. doi: 10.1016/s1389-1286(01)00216-X.
Y. Hu, l. Cao, F. Lv, S. yan, Y. Gong, & t. Huang. Action detection in complex scenes with spatial and
temporal ambiguities. En: IEEE 12th International Conference on Computer Vision, IEEE, 2009. doi: 10.1109/
ICCV.2009.5459153.
S. Liu, z. Song, GRAMO. Liu, C. Xu, h. Lu, & S. yan. Street-to-shop: Cross-scenario clothing retrieval via parts align-
ment and auxiliary set. En: Proceedings of IEEE Conference on Computer Vision and Pattern Recognition,
IEEE, 2012. doi:10.1109/CVPR.2012.6248071.

10. z. Haider, K.T.S. Alin, A. hussain, & A. hussain. Product associated displays in a shopping scenario.
En: Proceedings of the 4th IEEE / ACM International Symposium on Mixed and Augmented Reality, IEEE, 2005.
doi: 10.1109/ISMAR.2005.46.

11. h. Paulheim. Knowledge graph refinement: A survey of approaches and evaluation methods. Diario de

Web semántica 8(3) (2016), 489–508. doi: 10.3233/SW-160218.

12. G.L. Ji, k. Liu, S.Z. Él, & j. zhao. Knowledge graph completion with adaptive sparse transfer matrix. En:

Proceedings of the Thirtieth AAAI Conference on Artificial Intelligence, AAAI, 2016, páginas. 985–991.
JD.Com’s official website. Disponible en: https://www.jd.com/.

13.
14. Alibaba Group’s Buyer Official Forum. Disponible en: http://buyer.alibaba.com/forum.
15. METRO. Fu, C.M. Wong, h. Zhu, Y. Huang, Y. li, X. Zheng, j. Wu, C.M. Vong, & j. yan. DAliM: Machine learning
based intelligent lucky money determination for large-scale e-commerce businesses. En: Actas de la
16th International Conference on Service Oriented Computing, Saltador, 2018. doi: 10.1007/978-3-030-
03596-9_53.

16. Z.H. Kan, T.S. Alin, & MAMÁ. hussain. Price prediction of share market using artificial neural network (ANN).

International Journal of Computer Applications 22(2) (2011), 42–47. doi: 10.5120/2552-3497.

17. B. Krollner, B. Vanstone, & GRAMO. Finnie. Financial time series forecasting with machine learning techniques:
A survey. En: Proceedings of the 18th European Symposium on Artificial Neural Networks, 2010. páginas. 25–30.
Disponible en: http://works.bepress.com/gavin_finnie/37/.

18. S.N. Abdullah, & X. Zeng. Machine learning approach for crude oil price prediction with artificial neural
networks-quantitative (ANN-Q) modelo. En: Actas de la 2010 International Joint Conference on Neural
Networks (IJCNN), IEEE, 2010. doi: 10.1109/IJCNN.2010.5596602.

235

Data Intelligence

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
d
norte

t
/

a
r
t
i
C
mi
–
pag
d

F
/

1
3
2
2
4
6
8
3
7
7
4
d
norte
_
a
_
0
0
0
1
2
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Constructing a Scene-Based Knowledge System for E-Commerce Industries: Business Analysis
and Challenges

AUTHOR BIOGRAPHY

Min Fu is a Senior Engineer at Alibaba Group, Porcelana, and holds an honorary
fellowship with Macquarie University, Australia. He holds a PhD degree from
the University of New South Wales, Australia (UNSW) and has more than
four years of research experience and over six years of IT industry experience.
He has published over 23 publications in international conferences, incluido
the International Conference on Software Engineering (ICSE), the IEEE/IFIP
International Conference on Dependable Systems and Networks (DSN),
International Conference on Service Oriented Computing (ICSOC), y
journals including Journal of Software Practice and Experience (SPE) and IEEE
Transactions on Dependable and Secure Computing (TDSC). His main
research interests include service-oriented computing, cloud computing,
distributed systems, data mining, Big Data and machine learning, cyber
seguridad, software systems and architecture.

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

Qiang Chen is an Engineer II in the Knowledge Graphing Team of Alibaba
Group. He graduated from the Institute of Computing Technology of the
Chinese Academy of Sciences and is responsible for the project named Good
Guide. His research areas include natural language processing, machine
learning and knowledge graphing.

mi
d
tu
d
norte

t
/

a
r
t
i
C
mi
–
pag
d

F
/

1
3
2
2
4
6
8
3
7
7
4
d
norte
_
a
_
0
0
0
1
2
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Wei Lin is a Senior Engineer at Alibaba Group, Porcelana. He is currently a
part-time PhD student at the Chinese University of Science and Technology.
Prior to joining Alibaba Group, he worked as a researcher at the Keda Xunfei
Research Academy of China. His main research interests include knowledge
graphing, voice recognition and machine translation.

Data Intelligence

236

Constructing a Scene-Based Knowledge System for E-Commerce Industries: Business Analysis
and Challenges

Pei Wang is an Engineer II in Alibaba Group. He has worked as a research
engineer in the Institute for Infocomm Research in Singapore, which focuses
on advanced information technologies. Prior to joining Alibaba Group, él
was an algorithm engineer at the e-commerce firm, JD.Com Group. He has
rich experience in data mining in the area of e-commerce business. Él
currently works in text mining, e-commerce knowledge graphing and product
comments analysis.

Wei Zhang is a Senior Staff Engineer in Alibaba Group, China and is a master
supervisor in Fudan University, Porcelana. He holds a PhD degree from the
National University of Singapore and was once the head of the natural
language processing (NLP) application Lab of Institute for Infocomm Research,
Singapur. His works have been published in international conferences, semejante
as the Conference on Empirical Methods in Natural Language Processing
(EMNLP), ACM International Conference on Web Search and Data Mining
(WSDM), Conferencia AAAI sobre Inteligencia Artificial (AAAI), International Joint
Conference on Artificial Intelligence (IJCAI), and International World Wide
Web Conference (WWW). His main research interests include knowledge
graph, NLP, and machine learning. He is a standing reviewer for the journal
Transacciones de la Asociación de Lingüística Computacional (TACL) on NLP.