The Analysis of Intelligent Marketing Platform in High-Tech Products
by Data Mining Algorithm
CHUNG-CHIH LEE
Graduate School of Business and Operation Management
Chang Jung Christian University
TAIWAN
HSING-CHAU TSENG
Department of Business Administration
Chang Jung Christian University
TAIWAN
CHUN-CHU LIU
College of Continuing Education
Chang Jung Christian University
TAIWAN
HUEI-JENG CHOU
Department of Accounting and Information Systems
Chang Jung Christian University
TAIWAN
Abstract: - Under the background of the rapid development in internet technology, the whole marketing is
developing towards the direction of intelligence and high technology. The novel social network based on
internet technology occupies an important part of the marketing, and has also been widely concerned by the
academic community, because the internet makes information data transparent, and the mining, analysis and
algorithm research of a large amount of data can provide decision support for marketing and intelligent
marketing. Modern data mining analysis mode has become the main solution for data problems. With the
development of network technology, business intelligence related to the marketing of high-tech products will
become the key component of the future business system, which depends on the overall architecture of the
cloud and plays a core role in the process of data analysis and mining.
Key-Words: High-tech products, intelligent marketing platform, Data mining algorithm, Machine learning,
Cluster.
Received: July 24, 2021. Revised: January 5, 2022. Accepted: January 23, 2022. Published: February 9, 2022.
1 Introduction
Marketing is to convey to users a product, service,
brand value, the purpose is to sell and sell products,
services, improve the brand value [1]. Marketing
techniques include selecting a target market through
market analysis and segmentation, understanding
consumer behavior, and communicating the value of
the product to consumers through advertising. By
collecting and analyzing relevant information of the
company, intelligent marketing can accurately
identify market opportunities and formulate market
penetration strategies. Data analysis based on cloud
environment is an important analysis technology,
which depends on the cloud and the overall
architecture of the cloud [2]. The data construction
module plays an important role in the process of
data analysis and mining.
2 Data Mining Service Pattern
Analysis
2.1 Service Modeling Approach
In order to realize effective data analysis, it is
necessary to establish relatively effective models
based on data mining through cloud computing and
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2022.19.50
Chung-Chih Lee, Hsing-Chau Tseng,
Chun-Chu Liu, Huei-Jeng Chou
E-ISSN: 2224-2899
555
Volume 19, 2022
other relevant service technologies, describe and
identify functions through the whole process of data
service, and better analyze the relationship between
data and data. Generally, the data processing system
mainly has different levels of content. The purpose
of the infrastructure layer is to provide data
information for each work, access physical
resources through terminal interfaces, and provide
critical interfaces for the virtualization process. The
virtualization layer makes use of virtualization tools
to summarize different data resources in the cloud
environment, and logically encapsulates all the
resources, and provides them to the platform layer
for subsequent development from the allocation and
scheduling process. The platform layer is the key
part of the whole modeling system, and it is also the
service layer for data analysis. Finally, there is the
application layer, which directly provides services
to users. If users want to manage the service
structure through processing and requests, they need
to expand their identity authentication through the
application layer.
2.2 Data Mining Services
In the whole process of data mining, each step is
closely related to each other. The process of target
data analysis is completed by building a model, and
the contents of the model are evaluated and
discussed with the help of initial resources, and its
practical application methods in subsequent work
are analyzed. From this point of view, we carry out
content analysis according to the problems, so as to
understand the work objectives to be achieved and
master the requirements of sales objectives in the
field. If we evaluate the behavioral trend of
consumers, we can analyze whether the existing
resources can meet the needs of users through data
mining. If we can satisfy the relevant information,
we can further analyze the behavioral process of
target data mining. The whole process is divided
into several parts. The first is the initial data
preparation, because the data processing process is
not only for the large amount of data on the
network, but also to clean up the data of many
answers to determine how some overlapping
resources are allocated. In addition, in the process of
classification and integration of the basic data, more
valuable indicators need to be found from the
existing data to achieve the overall cleaning and
impurity removal of the data, and finally complete
the operation loading process. Then is the data
collection, data mining work, the premise is to
collect all the data, according to the data collection
process in the problem planning. For example, the
data included in multiple files or systems will
inevitably overlap, so it is necessary to perform
repetitive cleaning and unified storage management
of different data. In the face of a large amount of
data, how to select valuable contents according to
the actual needs of consumers can reduce a large
number of invalid work, reduce the work scale of
the calculated data, and select the appropriate tuples
to the greatest extent on the premise of keeping the
original data unchanged. In general, the filtering
process of data requires unified management and
control of the same type of data, especially how to
automate data processing in the context of massive
data, which has become the key to process standard
management and control. In this process also
involves the processing of the wrong data. After
clarifying the essence of the wrong data, the existing
defects are corrected. If there are a few errors in a
large amount of data, it will not affect the overall
degree of data perfection. On the contrary, if the
error ratio is too high, directly deleting these data
will inevitably affect the accuracy of the entire data
set, and then affect the subsequent operation.
Therefore, it is necessary to consider how to deal
with the null values in the data set, for example,
choosing the means of professional experience
analysis and regression analysis to compensate for
the null values. In the process of data conversion,
the data attributes are discretized into different types
of interval, if the data in the interval is mapped into
the discrete value of the response. According to the
whole process of data processing, the analysis
technology process is closely related to the data
processing of data sources. The correctness and
integrity of this data will directly affect the quality
of data mining. However, the current cloud
computing architecture has a strong computing
capacity, which can provide a large amount of data
analysis for the daily behaviors of enterprises, so as
to facilitate the analysis of relevant characteristics of
commodity attributes and user behavior tendencies.
The modeling process of data mining is a key part of
data analysis [3].
3 Task Learning Methods for
Intelligent Marketing in High-Tech
Products
Product sales time is the active cycle of products in
the market, and is also a key indicator to measure
the market development of related industries. From
the micro level, product sales time is not only highly
concerned by the seller, but also the main reference
for potential consumers to evaluate. Because in
general, the shorter the product sales time is, the
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2022.19.50
Chung-Chih Lee, Hsing-Chau Tseng,
Chun-Chu Liu, Huei-Jeng Chou
E-ISSN: 2224-2899
556
Volume 19, 2022
more popular the product is. From a macro
perspective, it can be used as a reference for market
liquidity. The shorter the product sales time is, the
better the market circulation performance of the
product is. Therefore, we study the data mining
methods related to product intelligent marketing,
and establish a decision system based on it. The key
is to determine the potential sales time of the
product. In the process of modeling and prediction,
it is necessary to analyze the possible adverse
effects of various potential factors. After obtaining
the characteristics from each type of data, the
regression method based on multi-task learning can
predict the time value. If necessary, it can carry out
experimental verification through a large number of
real product information, and determine the value
and role of multi-task learning method and decision
system through demonstration system. In countries
and regions with relatively developed market
economy, many enterprises begin to conduct in-
depth processing of business information through
data mining on the basis of the original information
system, so as to establish their own competitive
advantages and expand their turnover. American
express has a 5.4-billion-character database to track
its credit card business, which continues to be
updated as the business progresses. Express
company through the data mining, formulated the
"relevant settlement base (ship) to provide" sales
promotion strategy, that is to say, if a customer
bought in shops and a group of fashion, so in the
same shop to buy a pair of shoes, you can get a
bigger discount, increasing sales, also can increase
the utilization rate of expression. For example,
cardholders who live in London and recently flew to
Paris on a British airways flight might get a discount
card for a weekend trip to New York. In this way,
the data mining is very effective, which can not only
accurately locate the information of the target
customers, but also provide the necessary reference
for the performance work [4].
The main work. The main work of task learning
method can be divided into two aspects, one is the
marketing time of the product, the other is the
content of multi-task learning. Relevant experts and
scholars have conducted in-depth research on
product sales time, focusing on the relationship
between product marketing and product price. It is
worth mentioning that there are contradictions
among some research results. For example, many
documents mention that there is a positive
correlation between product marketing price and
time, while some studies believe that there is a
negative correlation between the two variables. The
reason for this kind of situation is that different
experimental Settings produce different results, for
example, different mathematical models will bring
different differentiation results to the variable
relationship. As researchers, we should pay more
attention to the attributes and information of the
commodity itself, so as to provide reliable solutions
for the platform construction. Specifically, each
method has its own advantages and disadvantages in
the prediction of results. In this study, we will try to
adopt machine learning technology to improve the
performance of the method, taking into account the
logarithmic changes involved in variables. Multiple
learning is a widely selected machine learning
method, which USES the correlation between tasks
to analyze the different performance of
classification regression. This method first appeared
in the research of neural network technology. In the
subsequent research, the multi-task learning mode
began to appear and formed a key research branch,
which is also the basis of the method adopted in this
project [5]. Its overall structure is divided into
cluster structure, graph structure and tree structure,
as shown in Fig 1, Fig 2 and Fig 3, respectively.
Fig. 1: Cluster structure
Fig. 2: The graph structure
Fig. 3: Tree structure
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2022.19.50
Chung-Chih Lee, Hsing-Chau Tseng,
Chun-Chu Liu, Huei-Jeng Chou
E-ISSN: 2224-2899
557
Volume 19, 2022
The difference between learning tasks is reflected
by the multi-task learning method. The main
difference lies in the hypothesis between different
types of task relationships, which is expressed in the
form of regular terms [6]. In part of the research
work, it is assumed that the tasks are related to each
other and are represented by low dimension. By
selecting some Shared features in the learning task,
the structure analysis in the case of multi-task can
be established and the learning task can be applied
to different fields. Characteristics analysis. Feature
analysis mainly focuses on the basic attributes of
goods, such as the price of goods and the nature of
goods. These attributes can be obtained directly
from the raw data. In the prediction model, all
numerical characteristics are standardized and
represented by discrete features. In general, if there
is no abnormal situation, this method is used to
carry out feature transformation. In order to further
improve the accuracy of prediction results of
commodity sales time, some meta-features should
be designed for integrated learning, and the
prediction results of other models should be trained
as characteristics. In the existing research, it has
been proposed that different models can be
combined to obtain more effective results. After
training the model with its own training data, the
predicted results can be added to the analysis
process of other contents as the original
characteristics. Intelligent platforms for high-tech
products should take into account certain historical
transaction information when the product is sold.
The main feature of high-tech products is that there
is a clear correlation between goods and technology
level. The higher the technology level, the more
difficult the management of products, and the
functions of products themselves also lead to
different types of product sales distribution. A more
reasonable solution is to establish different
prediction models for different types of products, so
that coefficient vectors clearly indicate the
dependence of commodities on influencing factors
under the premise of uniqueness. After we propose
to use multi-task learning for research, we can find a
middle way from different solutions and control the
similarity between models with the help of
regularization parameters. A multi-tasking approach
to product marketing time forecasting. The
prediction model of product marketing time
correlation should be discussed from the motivation
of multi-task learning method, and the advantages of
the model and algorithm should be analyzed. It is
important to note that meta-characteristics were not
added to the baseline approach in the study. In this
regard, we set up a system platform, which is used
to predict the sales time of commodities. Users can
input relevant information of commodities into the
system, and the system will give corresponding data.
If the data provided by the user is not complete, the
system platform will use the average as the default
value. This prediction model transmits information
to users by means of offline training, so as long as
users submit certain data, data results can be
obtained immediately. In this way, the system can
provide a very important ability to predict. The
potential marketing time of goods. Both sellers and
buyers can analyze the influencing factors under
different situation characteristics based on the
relevant content of products. The regression method
of multi-task learning also plays a significant role in
the prediction process and is of great value to the
purchasers of products.
4 Data Mining Algorithm under
High-Tech Product Marketing
Platform
4.1 Data Mining Process
In the marketing process, the key information lies in
the correlation between different commodities [7].
In order to analyze the attributes of commodities, it
is necessary to select the algorithm of association
rules, so that the resulting information can be
displayed by association rules to the greatest extent,
and the commercial value theory related to
commodities can be found from the mass of data.
After obtaining relevant information in a large
number of data sets, it is necessary to generate
different association rules according to the
relationship between the information. During the
association process, some non-redundant contents,
including some misleading rules, should be screened
out quickly. However, due to the existence of such
information, the quality of association rules we
obtain is not high, which may reduce the marketing
results of some products, thus affecting the quality
of product marketing. In order to improve this
aspect of the problem, it is necessary to effectively
improve the correlation between products and the
generation pattern, so as to avoid the negative
effects of useless rules. Generally, we can determine
several different types of information according to
the changes of users' interests. One is that the
objective degree of interest is expressed in the form
of data mining and information association. The
other is the degree of subjective interest, which is
used to evaluate the stability of rules and is also the
key to customer value analysis [8].
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2022.19.50
Chung-Chih Lee, Hsing-Chau Tseng,
Chun-Chu Liu, Huei-Jeng Chou
E-ISSN: 2224-2899
558
Volume 19, 2022
Customer value is mainly composed of current
value and potential value. Different values will be
driven by the will of consumers to produce different
values reflected by attributes, and information can
be obtained by means of estimation. In terms of the
analysis of current value, the information of
business format can be obtained through the sales
results of relevant products and platforms, and the
content can be extended according to the operation
characteristics and product sales capability, and
different scores can be made according to the sales
channels of some high-tech products. The difference
between platform marketing and general marketing
of high-tech products lies in that it is not affected by
geographical factors, but the degree of social
influence and consumers' consumption tendency
indirectly reflect their trust on products or
enterprises, thus affecting the subjective loyalty of
consumers. In this process, if we want to classify
customers, we should distinguish the difference
between different customers and realize the unified
management of customers. Consumer behavior and
psychological dynamics are helpful for enterprises
to provide personalized service contents in platform
marketing, fully explore customers' consumption
interests on the basis of existing data classification,
and identify the potential value of different
customers. For example, many users are inclined to
the performance of high-tech products, and the
performance of which type of products is suitable
for which customers can become the key to the
acquisition of preference attributes [9]. The details
can be shown in table 1 below.
Table 1. Customer value index analysis
Type of value
indicator
The specific content
The current
value
Business type information, customer
buying tendency, corporate social
impact, customer classification
The potential
value
Customer purchase volume, customer
consumption level, high-tech product
performance and attributes
4.2 Customer Interest Eigenvalue Acquisition
To realize intelligent sales of high-tech products, we
need to provide personalized key services for
different customers. In this process, we not only
need to accurately grasp the interests and trends of
different customers, but also need to analyze the
future work demand according to the marketing
situation of the existing products to carry out user
preferences and interest characteristics worth
acquiring based on the mass of data. There is a
concept involved in this work, namely "customer
portrait". Customer portrait in essence can be
understood as a kind of information label is a mark
of characteristics to the customer, the purpose is to
make enterprises understand the different habits and
customer information, such as they browse when
browsing the traces and commodity attribute
information can be provided the effective reference
data mining, and the customer data can be as a text
feature vector. After the behavioral data of different
customers are counted and classified, their
preferences can be obtained, and pure analysis can
be made on whether high-tech products are
applicable to users and whether they can obtain
high-tech products they are interested in by means
of interest measurement. Before this is done, the
first task is to preprocess the text file, i.e. to
accurately select the keywords that describe their
characteristics from the relevant information of the
customer. In the selection of general reference to
three aspects of the content, one is to express the
customer attributes of the words, that is, those who
have similarities and regularity of the description.
The second is to choose the description with specific
meaning, that is, the description that can reflect the
characteristics of things. The third is the hierarchical
distinction in the process of feature description
selection and the selection of different feature
contents in different information. Under the relevant
conditions of the same platform, different types of
consumers have different characteristics, and their
demands for products will be different according to
their personal tendency, thus generating the
classification in the text information. Good results
can be obtained by using the classification analysis
method of machine learning. After feature attributes
are extracted, the data collected by the platform are
manually modified to ensure the accuracy of
consumer classification information in the process
of continuous update. The similarity degree and
measurement method of various information can be
expressed by the following formula:
Cos= ï…
î¬¡ï…Ÿî¬¡î¬¡ï… î¬¡
(1)
a and b is the vector.  and  is the the length
of a and b, respectively.
It can be seen that the cosine value is used in the
formula to analyse the similarity degree of different
information. Information a and information b are
regarded as two different information. The closer
the value of two vectors is to 1, the higher the
similarity degree of information is. Conversely, the
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2022.19.50
Chung-Chih Lee, Hsing-Chau Tseng,
Chun-Chu Liu, Huei-Jeng Chou
E-ISSN: 2224-2899
559
Volume 19, 2022
closer the cosine value is to 0, the lower the
similarity between the two information is. For the
marketing platform, these two pieces of information
have a clear reference value, because the more
characteristic elements contained in the two pieces
of information, the higher the degree of relevance
and relevance of the two consumers are potential
customers, which is conducive to the marketing of
similar products. Of course, there are different types
of products in the market at this stage, and even
personalized recommendation of customer interest
will be affected by a variety of factors. Enterprises
face different customers and can also consider
modelling based on different attributes of the goods.
For a customer who needs to buy products, the
degree of correlation between different products will
affect their consumption behaviour. For example, if
a high-tech electronic product is of interest to a
certain customer, the products of other brands that
are close to such products will also make the
customer interested. At this time, we can choose
representative description of commodity
characteristics to calculate the degree of similarity
between the two commodities according to different
characteristics of the products. During the operation
of the platform system, these modules will read by
historical information and obtain the user's interest
value. After the content information in the page has
been classified, different information categories can
be added to the user evaluation of the interest file. If
there is something similar to this property in a
traditional dataset, the feature can be added directly
to the original dataset, or a new dataset can be
created in the file. Product preference analysis. In
the previous paper, the customer's interest and
purchase desire have been determined according to
different types of product attributes. After obtaining
the customer's basic information and adding these
attributes into the model research, the correlation
can be regarded as the prediction result of product
preference data. All of these contents can be
expressed as the customer's interest category in the
model, and then the data of weight value is arranged
to obtain the customer's preference interest matrix
for the product. In fact, for the same type of high-
tech products, there will be different categories, and
accurate positioning of the correlation between
product attributes and customer interests will
provide us with key information, so as to facilitate
the acquisition of result set data. Overall, under the
big data environment, customer similarities of
association rules based on the system to exploit their
interest trend, analysis of customer value and
product value under different situation, with the aid
of clustering algorithm in data mining modelling
classification, and on this basis, evaluate customer
characteristic vector and information between the
product attributes [10].
5 Application of Marketing Platform
5.1 Analysis of Enterprise Marketing Data
The purpose of the analysis of enterprise marketing
data is to conduct data mining for the overall
manager of the enterprise, and obtain valuable
information from it, which is helpful to analyze the
product, the operation situation in the market and
the trend of market changes, and to develop
personalized intelligent marketing strategies for
different customers. Generally speaking, the way of
data presentation is described by using map
information. The color of different depth represents
the difference of product sales volume, which can
also provide effective reference for the selection of
marketing strategy. In the whole process of
marketing data analysis, the importance of
customers is obvious. To measure the potential
value of a customer to an enterprise, it is necessary
to analyze the relationship between customer value
and sales according to the relevant theories of life
cycle, so as to locate the consumer behavior of
customers. For example, the development potential
of customers is mainly reflected in their ability to
purchase commodities. The higher the customer is,
the more capital he has to buy a certain product.
Especially for intelligent products, the promotion
mode of products is obviously more important,
because it can clarify the purchase philosophy of
different customers, take the initiative to promote a
large number of certain products, with the help of
data mining algorithm to integrate information into
the platform, so that enterprises can make targeted
product recommendations according to individual
users.
5.2 Marketing Platform
Within the marketing platform, the information
of all customers can be identified and
processed by the application layer and cloud
computing platform. Both the online and
offline shelf of products and the analysis of
customers' purchasing behaviour can be
adjusted through the platform. When customers
visit the platform, they will generate service
information, which will become the reference
basis of the enterprise system after
classification, and enable the enterprise to find
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2022.19.50
Chung-Chih Lee, Hsing-Chau Tseng,
Chun-Chu Liu, Huei-Jeng Chou
E-ISSN: 2224-2899
560
Volume 19, 2022
similarities among different customers to meet
the actual needs of the enterprise for product
marketing business. It is foreseeable that this
personalized recommendation model will also
become a major development trend in the
future marketing.
6 Conclusion
Nowadays, the intelligent marketing mode provides
a clear direction for the development of enterprises,
and the personalized marketing system also provides
a reference for the sustainable development of
enterprises. In this study, data mining algorithm and
hybrid model are used to improve the personalized
recommendation of products. Of course, in a sense,
the role of data mining will be paid more attention
in the future research, because it conforms to the
development trend of modern intelligent marketing
and the development law of future business
operation. There are still some deficiencies in our
research, which need to be further improved in the
future practice. For example, in the case of customer
classification, it needs to be more detailed, select a
wider range of regions for personalized
recommendation, according to the customer's
interest trend to optimize the operation and sales
model of the enterprise, to create greater value of
profits. With the support of intelligent platform, the
positioning of customers will be more accurate,
which will help enterprises to make business
operation plans, maintain their sustainable
development by good economic means, and
promote the modern marketing of products and the
smart marketing of products.
In the future, relative scholar can try to implement
decision support system based on proposed
intelligent marketing platform for enterprise to use.
References:
[1] Suraksha Gupta, David Gallear, John Rudd,
PanteaForoudi, (2020). The impact of brand
value on brand competitiveness. Journal of
Business Research, Vol. 112, 210-222.
[2] Omar M. Ahmed, Lailan M. Haji, Hanan M.
Shukur, Rizgar R. Zebari, Shakir M. Abas,
Mohammad A. M. Sadeeq. (2020). Comparison
among cloud technologies and cloud
performance. Journal of Applied Science and
Technology Trends, Vol. 1, No, 2, 40-47.
[3] Zhang yanqin, Pan liqiang. Application of
Apriori algorithm in WEB book sales
intelligent system, Network security technology
and application, Vol. 9, 2017, pp. 58-59.
[4] Liu ping, Fan jinding, Sun jianhua, Yang
ruogu, Song jian. Research on SmartC on Petro
oil and gas production big data mining system
technology based on intelligent oil and gas field
cloud platform, Digital design, Vol. 5, No. 03,
2016, pp. 47-52.
[5] Nan nan. Application research of improved
Apriori algorithm in solar energy sales system,
Journal of tongling vocational and technical
college, Vol. 14, No. 4, 2015, pp. 43-45.
[6] Zhang lei, Han wenteng, Wang guofei,
Customer quality rating in the telemarketing
mode of insurance industry (based on data
mining), Enterprise leader, Vol. 22, 2015, pp.
107 - 109.
[7] Mu jun. Discussion on business intelligence
architecture and data mining technology based
on industrial chain collaboration platform.
Journal of jixi university, Vol. 15, No. 5, 2015,
pp. 25-28.
[8] Lin hui. Research on accurate marketing of
local retail industry in the era of big data --
from the perspective of data mining, Business
times, Vol. 04, 2014, pp. 58-59.
[9] Cao qiuhong. Research on traffic intelligent
data mining -- storage and centralized
application of data based on unified signaling
monitoring platform, China new
communications, Vol. 15, No. 17, 2013, pp. 66.
[10] Gao pan, Guo li. Application of multi-strategy
data mining model in book sales intelligent
analysis system. Information systems
engineering, Vol. 4, 2010, pp. 60.
Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en
_US
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2022.19.50
Chung-Chih Lee, Hsing-Chau Tseng,
Chun-Chu Liu, Huei-Jeng Chou
E-ISSN: 2224-2899
561
Volume 19, 2022
