Iee Digital Marketing

119
Digital Marketing Strategies Leveraging Data Fusion and

Communication Technology for Effective Human Resource
Management and Organizational Configuration
Junxin Chang1* , Shuo Wang2 and Quan Li3
1,2,3Training center of State Grid Hebei Electric Power Co., Ltd, Shijiazhuang 050031, Hebei, China,
[email protected],[email protected],[email protected]
Corresponding author: Junxin Chang,[email protected]
Abstract. In the new normal economic development environment, strengthening

human resource management to obtain core competitive advantages is of key
significance for enterprises to cope with market competition and obtain economic
benefits. Effective project management is an important way to improve the
competitiveness of enterprises, and its core content is project planning, scheduling,
and control. Project scheduling is to study and solve how to arrange resources
according to time according to various tasks of the project process, so that the
predetermined optimization goal can be realized. We propose a memetic algorithm
for the classical resource-constrained project scheduling problem. Memetic uses the
task linked list as the individual code, the single-point crossover operator as the
recombination operator and applies different combinations of five local search
processes to improve the offspring individuals. Many cases from three groups of case
sets J30, J60 and J120 in RCPSP's standard question library PSPLIB are used to test
MA, and the results show that MA has strong competitiveness.
Keywords: project planning;human resource management;Digital Marketing

Strategies;project scheduling,
DOI: https://doi.org/10.14733/cadaps.2024.S4.119-134
1 INTRODUCTION
With the changes of the globalized economic development situation, enterprises are facing more and
more complex external environment. At this time, strengthening human resource management will
help enterprises to improve their risk resistance ability, reduce brain drain, and ensure operational
efficiency [16]. Under the new situation, the human resource management of enterprises is more
systematic and comprehensive, and it is necessary to adjust the strategic plan in a timely manner
Computer-Aided Design & Applications, 21(S4), 2024, 119-134

© 2024 CAD Solutions, LLC, http://www.cad-journal.net
120
in combination with changes in the external environment and internal business conditions, to play a
more efficient human resource management function [7]. At the same time, human resource
management in the new era places more emphasis on the participation of all employees and the
construction of the whole process. The initiative and autonomy of human resource management in
various departments should be improved to better stimulate the vitality and creativity of talents and
enhance the contribution of human capital to promoting economic benefits. important role. Under
the complex and changeable economic situation, enterprises should update human resource
management concepts, promote organizational reform, and improve organizational training and
performance appraisal systems [21]. At present, the lack of scientific in the management concept,
system construction, performance appraisal, and information construction of enterprises makes
human resource management activities unable to effectively help enterprises to enhance their core
competitive advantages, and seriously affects the stability of their internal operations [2]. Therefore,
enterprises should comprehensively analyze the current problems in human resource management,
and then build an efficient, dynamic, and flexible human resource management system, and
formulate relevant human resource management risk response mechanisms to create an
organizational culture that is conducive to ensuring personnel stability. Institutional system to ensure
the economic performance of enterprises.
The name Project Management first appeared in the Manhattan Project in the United States. In
the early 1960s, famous mathematicians in my country introduced the idea of project management
into China, which had a positive impact on the management of related fields in my country. Project
management includes project time management, cost management and quality management.
Project scheduling is closely related to project time and cost and is one of the core contents of project
management [24]. Early project management did not consider resource constraints, but in recent
decades, people have increasingly realized that the project scheduling plan prepared using the
classical project scheduling methods CPM and PERT often cannot be implemented smoothly under
the condition of limited resources. The research on RCPSP can be summarized into two aspects: one
is to establish RCPSP models that meet different requirements, and the other is to solve RCPSP
algorithms. RCPSP is mostly NP. Hard problems are more difficult to solve. This paper proposes a
Memetic algorithm (MA) for RCPSP. MA uses the task linked list as the individual code, the single-
point crossover operator as the recombination operator and uses different combinations of five local
search processes to improve the offspring individuals [6]. Many cases from the three groups of case
sets J30, J60 and J120 in the RCPSP standard problem library PSPLIB are used to test MA, and the
results show that MA has strong competitiveness. In the whole process, relying on effective data
fusion analysis and the communication of individual codes in the task chain, our method provides
effective guidance information for subsequent human resource allocation.
2 RELATED WORK
2.1 Problems Existing in Human Resource Management

First, the human resource management system of the enterprise is not perfect, the connection and
cooperation of each module are unreasonable, and there is a lack of integrity and systematization
[8]. When enterprises recruit talents, they ignore the combination with the follow-up talent training
mechanism and talent allocation mechanism, which leads to the mismatch between talents and job
requirements, and it is difficult to attract talents through training and promotion channels [3]. At
the same time, the formulation of the compensation incentive system of the enterprise is not
scientific and reasonable enough, it is difficult to fully mobilize the enthusiasm and subjective
initiative of talent work, and it fails to cultivate the sense of belonging of talents, which makes it
difficult for human resource management to improve economic benefits [25]. In addition, the
company ignores the follow-up performance feedback when carrying out human resource

121
management activities, so that the work problems and business risks generated by personnel at all
levels cannot be solved and fed back in time; secondly, it is difficult for the company to formulate
human resource strategies to meet development needs. , the investment in human capital is
insufficient, and it is impossible to create a talent management system that can attract employees,
thereby affecting the acquisition of corporate economic benefits [22]. Some enterprises lack
scientific human resource management concepts, pay too much attention to short-term human cost
control, while ignoring long-term human capital investment returns, and pay insufficient attention
to talent training and salary incentives, which is prone to brain drain, which in turn causes
management problems. confusion, affecting the acquisition of the ultimate economic benefits [9].
And some enterprises only focus on short-term labor cost control and lack a correct understanding
of how to accumulate human resources to obtain higher levels of economic performance; finally,
enterprises lack a scientific talent allocation mechanism, resulting in an unreasonable human capital
structure and affecting various business management activities. successful implementation [1].
Talent allocation is related to the combination of people and positions, the cultivation of subsequent
talents, and how to retain the talents of the enterprise. However, the lack of scientific talent
development planning in the enterprise makes the enthusiasm of some employees not high, thus
restricting the innovation and profit growth of the enterprise.Lastly, digital marketing can enable
effective talent development planning and performance feedback mechanisms. Learning
management systems and online training programs can be implemented to provide continuous
learning and skill development opportunities for employees. Moreover, digital tools can facilitate
regular performance evaluations, feedback loops, and goal tracking, enabling timely recognition,
support, and corrective actions.
2.2 The Role of Human Resource Management

By improving and innovating human resource management models, enterprises can help them
achieve their strategic goals, and can help them find problems such as labor costs, human resource
allocation, and brain drain in a timely manner, thereby improving the overall organizational
management efficiency of the enterprise. Enterprises build a complete human resource management
system, which is helpful to comprehensively control the use of human resources in various
departments and improve the standardization of human resource allocation [19]. At the same time,
the enterprise builds a human resource management system involving all employees, which can
timely discover and feedback human resource problems and risks to various departments, to help
each department take effective measures to improve the performance appraisal process, personnel
recruitment process, organization training process, etc. Through the formulation of human resources
strategy, the enterprise can ensure the future talent selection, talent allocation and talent cultivation,
control the risk of future personnel management in advance, reduce the phenomenon of
unreasonable personnel allocation to unreasonable positions, and effectively control the enterprise
Unnecessary waste of human resources [13]. Specifically, only by setting up a scientific human
development strategy can an enterprise improve personnel satisfaction, protect human capital,
reduce the negative impact of personnel turnover on the economic benefits of the enterprise, help
employees find performance gaps, and promote employees to improve their work efficiency. Reduce
work errors [17]. Moreover, when the enterprise formulates reasonable and fair compensation, it
can enhance the loyalty of employees, reduce the turnover intention of employees, and promote
employees to obtain high performance by improving their own ability level [4]. In addition, under
the drastic changes in the external policy environment and market environment, enterprises should
formulate scientific employee training plans to improve employees’ understanding of external
policies and market conditions, update their ideas and knowledge and skills in a timely manner, and
keep up with the development of the times [11]. Talent is the premise of enterprise development,
but at the same time, it also generates a lot of labor costs. Only by improving the efficiency of
human recruitment, ensuring the performance of human benefits, improving the efficiency of human

122
resource allocation, and meeting the needs of talents in various business management activities,
can enterprises pass the human capital. Continue to leverage the economic benefits of more
enterprises [20].
2.3 Resource Constrained Project Scheduling Problem

A general RCPSP can be described: a project contains a task, task l is the only task that starts the
earliest, and task is the only task that finishes the latest, all of which are virtual tasks (no resource
consumption and execution time is 0), represent the beginning and end of the entire project,
respectively [10]. The task must choose one of M execution modes for execution (each execution
mode corresponds to certain resource requirements and execution time), and the execution mode
cannot be interrupted or changed during the execution process. Executing task J in the MTh
(1≤m≤Mj) mode requires the kth (seven = 1...., K) type of updateable resources to be rPjr, k, which
requires the kth (row = 1...., IV))is an amount of non-updatable resources, and the execution time
is time [15]. Virtual task I and, have only one execution mode. If the project starts from phase 0,
the upper limit of the project duration is D [12]. During the entire project duration, the available
amount of the kth renewable resource in each stage is constant RPk (7=1…., K), and the total
amount of the nth non-renewable resource is such as (gate=1…., IV). In addition to resource
constraints, due to technical or project organization reasons, there are also time constraints between
tasks, and this time constraint relationship is related to the execution mode selected by related tasks
[5]. When task i chooses the mjth mode to execute, and task/selects the first mode to execute, the
start time S of task i is related to the task, and the start time S needs to satisfy the constraint of
formula (1),
SSmmin
imj
 STj − STi  SSmmax
im j
(1)
The SF, FS, and FF-type time constraints between tasks can also be described by Equation (1).
Solving RCPSP is to determine the execution mode and start execution time of each task under the
condition of ensuring resource constraints and time constraints between tasks, so that one or more
project indicators are optimal.
3 METHODOLOGY
3.1 Memetic Algorithm

Memetic algorithm (Memetic Algorithm, referred to as MA) is a swarm intelligence optimization
algorithm based on human cultural evolution strategy. The term Meme was first proposed by a
British scholar in 1976. In his paper "The Selfish Gene", Meme is used to represent the information
unit transmitted by people during communication, which is translated as "cultural gene" or "cultural
genetic factor". Meme will change due to personal thoughts and understanding in the process of
dissemination, so the information passed from the parent to the child can be changed, which is
manifested in the process of local search in the algorithm. In 1992, the Memetic algorithm was
formally proposed and applied to the optimal solution of the traveling salesman problem. In 1994,
he introduced the Memetic algorithm in detail in the paper (Formal memetic algorithms), discussed
the optimization mechanism of the algorithm, and focused on the local search strategy of the
algorithm. Since then, the Memetic algorithm has attracted the attention of many scholars, and the
research content and application fields have also been continuously expanded and deepened. Like
the genetic algorithm that simulates the biological evolution process, MA is an algorithm that
simulates the cultural evolution process. The mechanism of combining global search and local search
of MA makes its search efficiency higher than that of traditional GA in some problems and is widely
used in solving complex problems. In essence, MA proposes a framework and a concept. Under this
123
framework, different local search methods are used to form different algorithms. For example, the
global search strategy can use genetic algorithm, evolution strategy, etc. The strategy can use
simulated annealing, tabu search, greedy algorithm, guided local search, etc. The execution process
of MA is to initialize the population first, generate a set of chromosomes with reasonable spatial
distribution (random solution), and then obtain the optimal solution (near-optimal solution) through
local search. In each iteration, chromosomes are updated through crossover, mutation, and local
search. The algorithm flow is shown in Figure 1.
start
Initialize
population
Whether the stop Output

YES
conditions are met results
NO
end
Evaluation and
selection
Cross Ten Local

Search
Mutation+local
search
Figure 1: Memetic Algorithm Flow.
It can be seen from the flow of the algorithm that IvlA fully absorbs the advantages of GA and LS.
MA not only has a strong global optimization ability, but also performs Local Search after each
crossover and mutation and eliminates bad individuals as soon as possible. By optimizing the
population distribution, it reduces the number of iterations, improves the solution efficiency, and
ensures the quality of the output solution. MA combines the genetic algorithm with the local search
process. From the perspective of genetic algorithm, MA is a standard genetic algorithm that runs on
the local optimal solution. Compared with the standard genetic algorithm, its efficiency is mainly
reflected in the reduction of the search space. From the point of view of local search algorithm, MA
is a local search algorithm starting from multiple points. Compared with the local search algorithm
starting from multiple points, the efficiency of MA can be explained as starting from multiple better
solutions for local search., the optimal solution (near-optimal solution) can usually be reached after
less iterations.
3.2 Solving the MA of RCPSP

In this paper, MA uses the task linked list satisfying the priority relationship as the individual
encoding and uses the serial scheduling scheme to decode. When using the meta-heuristic algorithm
to solve the RCPSP, the operator usually acts on the encoding of the individual rather than the
problem solution itself, and then obtains a task scheduling scheme that meets the requirements by
invoking the appropriate encoding and decoding processes. For decades, many scholars have
proposed many meta-heuristic algorithms for RCPSP, and these algorithms are worthy of reference
124
in the process of designing RCPSP solving algorithms. The encoding methods of the meta-heuristic
algorithm for solving RCPSP are classified into five kinds: task linked list encoding, task priority value
linked list encoding, scheduling rule linked list encoding, shift vector linked list encoding and
scheduling scheme linked list encoding. According to the research conclusion, the meta-heuristic
algorithm using task linked list encoding and serial scheduling scheme (SSS) decoding is better than
other encoding methods. Therefore, the MA proposed in this paper is encoded using a linked list of
tasks and decoded using the SSS process.
3.3 Sorting Of Non-Dominated Sets

In 1994, a non-dominated sorting method based on sorting selection and niche technology was
proposed, and a non-dominated genetic algorithm (NSGA) for solving multi-objective optimization
problems was proposed combined with genetic algorithm. After that, in 2002, the calculation of the
non-dominated ranking level in NSGA and the fitness sharing strategy that needs to set the sharing
radius were improved, and NSGA-II was proposed. The calculation time complexity of the non-
dominated ranking level in NSGA-II is lower., and the calculation of crowding degree distance does
not need to set parameters manually, this paper adopts the method of sorting non-dominated sets
in NSGA-II to calculate the level and crowding degree distance of everyone. The selection of
individuals in MOCS is based on the comparison between individual individuals, that is, for each
individual j nest in the population nest, an individual ' i nest is randomly selected in the newly
generated population ' nest, if ' i nest dominates j nest, then Replace j nest with ' i nest. In IMOCS,
combine nest and ' nest to form a population tempnest containing 2n individuals, sort tempnest as
a non-dominated set, and add two attributes rank i and distance i to each individual i in tempest.
Select n individuals from tempnest, first select individuals with lower levels, and if the levels are the
same, select individuals with greater crowding.
4 EXPERIMENTS
4.1 Evaluation Indicators

Two performance evaluation indicators are used in the experiment: the generalized distance GD
(Generational Distance) and the diversity of the Pareto frontier (* P). GD is used to measure the
closeness of the desired *P to the real frontier P. The smaller the GD, the closer the desired *P is to
P. The definition of GD is as follows:
1/ p
1  
P*
GD = *  di p
(2)
P  i =1 
 
Among them, id is the Euclidean distance between the ith individual and the corresponding real
Pareto frontier, p is the number of objectives in the optimization problem, and for the optimization
problem of two objectives, p = 2. Population diversity is defined as formula (3):
 p
1
P* −1 
= p   d m +  di − d
e
 (3)
 
 dm + P d  m=1
e
m =1
* i =1

where it is the Euclidean distance between two adjacent points in *P, d is the average of all ids, and
em d is the sum of the boundary solution of *P found on the MTh target and the extreme solution

125
of the true Pareto frontier P. distance between. Considering both the breadth and uniformity of the
required *P distribution, em d is used to measure the breadth of the *P distribution, and || id d is
used to measure the uniformity of the *P distribution. For a set of ideal Pareto frontiers (uniform
distribution id d, em d =0), =0, therefore, the smaller is, the better the diversity of the Pareto
frontier is. Dimension d = 30 in test cases ZDT1, ZDT2, ZDT3, d = 10 in ZDT4, and d = 5 in LZ. In
the IMOCS algorithm, it is found that the probability 0.5 a p is the same as that of MOCS. The initial
value of dynamic adaptive change 0 of SCH, ZDT1, ZDT2, and ZDT3 is 0.1, the change rate K=1.07,
the threshold value T=0.3, and 0 is limited to [0.01,2]; ZDT4, the initial value of the dynamic
adaptive change 0 of LZ is 0.5, the change rate K=1.07, the threshold value T=0.15, and 0 is limited
to [0.1,5]. The settings of the above parameters, the limited range of 0, the rate of change K, and
the threshold T are obtained from multiple experiments.
4.2 Non-Dominated Set Sorting Performance Analysis

In this paper, the IMOCS algorithm is programmed and implemented in the computing environment
of Pentium(R) Dual-Core T4300 @2.10GHz CPU, 2GB RAM, Windows 7, and MATLAB 2010b. In the
experiment, the randomness of the data will have a certain impact on the test results. To reduce
this impact and reflect the performance of the algorithm more accurately, each algorithm is run
independently for 10 times in the test, and then the average and variance of the GD sum are
calculated. The effect of step size control on IMOCS performance. To verify the influence of the step
size control variable 0 on the performance of IMOCS, this paper conducts comparative tests on the
performance of IMOCS when 0 is fixed (0 = 0.01) and 0 is dynamically adaptively changed. Setting
n=200, Mixite=100, Table 1 gives the mean GD and variance 2 of GD between *P and P for 6 test
instances, and Table 2 gives the mean and variance 2 of *P. It can be seen from Table 1 that the
dynamic adaptive change of 0 makes the IMOCS have better performance than the fixed value, and
the GD of the Pareto frontier of all test cases is reduced compared with the fixed value of 0, that is,
when the dynamic value of 0 in the IMOCS is 0 When adaptively changing, the desired Pareto front
is closer to the real Pareto front. And it can be seen from Table 2 that when 0 in IMOCS changes
dynamically and adaptively, the diversity of the Pareto fronts it seeks is also better.
algorithm SCH ZDT1 ZDT2 ZDT3 ZDT4 LZ

GD  2
GD  2
GD  2
GD  2
GD  2
GD 2
IMOCS/fixe 5.72 8.45 1.83 4.50 1.64 3.97 6.47 1.45 1.37 5.25 3.72 5.08
d 0 E-05 E-11 E-05 E-08 E-05 E-08 E-06 E-08 E-02 E-04 E-04 E-06
IMOCS/dyna 3.79 5.37 8.06 5.66 9.99 8.47 1.04 1.67 1.23 1.26 1.68 2.44
mic 0 E-06 E-11 E-07 E-12 E-07 E-12 E-06 E-13 E-06 E-12 E-04 E-07
Table 1: Mean value and variance of GD obtained when step size control variable is dynamically
adaptive and fixed.
Figure 2 presents the Pareto frontier for each test case. For ZDT4, the IMOCS under the dynamic
adaptive change is 4 orders of magnitude smaller than that required by the fixed IMOCS, which is a
great improvement, because the fixed IMOCS fails to search the global Pareto front in 10
independent runs, and the IMOCS under dynamic adaptive changes all search for the global Pareto
frontier, so this is more intuitively reflected in the Pareto frontier diagram of ZDT4.
algorithm SCH ZDT1 ZDT2 ZDT3 ZDT4 LZ

126
 2  2  2  2  2  2
IMOCS/fixe 5.15 1.75 5.87 1.12 5.66 9.85 7.06 1.18 7.66 6.28 1.41 1.69
d 0 E- E- E- E- E- E- E- E- E- E- E- E-
02 05 02 04 02 05 02 04 02 04 00E 03
IMOCS/dyna 4.66 2.14 5.14 7.73 5.15 2.38 6.88 1.74 6.03 3.18 9.62 2.26
mic 0 E- E- E- E- E- E- E- E- E- E- E-02 E-
02 05 02 04 02 03 02 04 02 04 02
Table 2: The mean value and variance of diversity obtained when the step size control variable is
dynamically adaptive and fixed.
Figure 2: Pareto Frontier of Test Cases.
Figure 3 presents the GD convergence curves for the 6 test cases. For the SCH with only one-
dimensional variables, the two types of IMOCS under 0 converge very quickly; for ZDT1, ZDT2,
ZDT3, the IMOCS under the dynamic adaptive change 0 is significantly faster than when 0 is fixed;
for ZDT4, at 30 times Before the iteration, the GD curve obtained by IMOCS under dynamic adaptive
127
change 0 is above the GD curve obtained by IMOCS under fixed 0. This is mainly because the IMOCS
under dynamic adaptive change 0 is in the exploration period, and the change range of 0 is large.
After 30 iterations, the GD curve obtained by IMOCS under the dynamic adaptive change of 0 is
below the GD curve obtained by IMOCS under fixed 0, and the final magnitude of GD is e-5, which
is reflected in Fig. 2(e) as The IMOCS under dynamic adaptive change 0 reaches the global Pareto
front, while the IMOCS under fixed 0 falls into the local Pareto front, and its final GD order of
magnitude is e-2; it can be seen from the convergence curve of LZ that under fixed 0 There are
several obvious fluctuations in the GD value of the IMOCS in the search, while the IMOCS under the
dynamic adaptive change 0 is relatively stable. The dynamic adaptive change of 0 can make the
IMOCS converge quickly and stably.
Figure 3: GD Convergence Curve of Test Cases.
Figure 4 shows the change curve of 0 of dynamic adaptive change during 1000 iterations, and its
test example is ZDT4. It can be seen from Figure 4 that before 40 iterations, the value of 0 fluctuates
greatly, which is an adaptive adjustment process of 0. The algorithm is in the exploration period,
and the value of GD decreases rapidly during this period. The convergence curve of GD is shown in

128
Figure 3(e); from 40 iterations to 70 iterations, the value of 0 is stable at around 1.3, and the
algorithm is in the convergence period, during which the value of GD slowly decreases to the order
of magnitude e -4 Nearby, it indicates that the search reaches the global Pareto front, which is shown
in Figure 2(e) as the IMOCS under dynamic adaptive change 0 reaches the global Pareto front, while
the IMOCS under fixed 0 falls into the local Pareto front; after 70 iterations, The value of 0 gradually
decreases, and becomes the set minimum value of 0.1 after 120 times. The algorithm has entered
a stagnation period. After that, the value of GD stabilizes around the order of magnitude e -5, and
there is no obvious change.
Figure 4: Variation Curve of an in ZDT4.
Performance comparison analysis of IMOCS and MOCS. To verify the effectiveness of the IMOCS
algorithm, this paper compares its performance with MOCS. The GD values of five test instances of
MOCS at n=50, Mixite=500 is given. In this paper, the GD values of IMOCS under the same
conditions are calculated, and the experimental results of the two algorithms are listed in Table 3.
It can be seen from the table that the GD values required by IMOCS are all smaller than MOCS,
indicating that the performance of IMOCS is better than that of MOCS.
algorithm SCH ZDT1 ZDT2 ZDT3 LZ

MOCS 1.25E-08 1.17E-08 2.23E-06 2.88E-06 4.16E-06
IMOCS/dynamic
0 7.69E-08 1.09E-08 1.47E-08 3.42E-08 3.16E-06
Table 3: GD Values Calculated by IMOCS and MOCS.
The test results of 4 test instances of NSPSO at n=200 and Mixite=100 are given, and the test
results of IMOCS under the same conditions are given in this paper. Table 4 shows the mean and
variance 2 of GD between *P and P for the 4 test instances, and Table 5 shows the mean and
variance 2 of *P. It can be seen from Table 4 that for the ZDT series, the GD obtained by IMOCS is
1-2 orders of magnitude smaller than that of NSPSO, which means that the Pareto front obtained by
129
IMOCS is closer to the real Pareto front than that obtained by NSPSO, and its solution deviation is
smaller. IMOCS solution is more stable. It can be seen from Table 5 that the requirements of IMOCS
are all smaller than those of NSPSO, which indicates that the diversity of Pareto fronts required by
IMOCS is better, and its deviation is smaller, which further shows the stability of IMOCS.
algorithm ZDT1 ZDT2 ZDT3 ZDT4

GD  2
GD  2
GD  2
GD 2
NSPSO 7.78 8.14 8.06 3.06 3.41 2.55 7.83 6.92
E-05 E-06 E-05 E-06 E-04 E-05 E-05 E-06
IMOCS/dynamic
0 8.06 5.66 9.99 8.47 1.04 1.67 1.23 1.26
E-07 E-12 E-07 E-12 E-06 E-13 E-06 E-12
Table 4: GD mean, and variance calculated by IMOCS and NSPSO.
algorithm ZDT1 ZDT2 ZDT3 ZDT4

  2
  2
  2
 2
NSPSO 7.68 3.01 7.59 2.78 9.15 3.94 7.69 3.58
E-02 E-03 E-02 E-03 E-02 E-03 E-02 E-03
IMOCS/dynamic
0 5.14 7.73 5.15 2.38 6.88 1.74 6.03 3.18
E-02 E-04 E-02 E-03 E-02 E-04 E-02 E-04
Table 5: Mean and Variance of Diversity Calculated by IMOCS and NSPSO.
Its performance was compared with that of NSGA-II. Deb improved NSGA in 2002 and proposed
NSGA-II but did not give an experimental analysis. The GD values of five test instances of NSGA-II
at n=50 and Mixite=500 is given. In this paper, the GD values of IMOCS under the same conditions
are calculated, and the experimental results of the two algorithms are listed in Table 6. It can be
seen from Table 6 that the GD values required by IMOCS are all smaller than NSGA-II, indicating
that the performance of IMOCS is better than that of NSGA-II.
algorithm SCH ZDT1 ZDT2 ZDT3 LZ

NSGA-II 5.73E-04 3.33E-03 7.24E-03 1.14E-02 4.19E-03
IMOCS/dynamic
0 7.69E-08 1.09E-08 1.47E-08 3.43E-08 3.16E-06
Table 6: GD Values Calculated by IMOCS and NSGA.
4.3 MA Performance Analysis

To determine the algorithm parameter values of MA, the author firstly conducted some preliminary
experiments. In the preliminary experiment, the MA solves the first case under each seed in a set
of J120 cases (ie cases J1201 1.SM, J1202-1.SM, J12060_1.SM). Given that LSHY'BRID embeds
other 4 local search procedures, to reduce the computational complexity, only the LSHYBRID local
search procedure is applied in the preliminary experiments. First, to test the crossover probability
P. Influence on MA performance, take pop Size=80, Gen=30, mixite=20, P. = {0.75, O. 80, O. 85,
0.90), run MA to solve each problem case 4 times, and set different hoist values each time. The
calculation results are shown in Table 7. The column label "Sum" in Table 7 represents the total
duration of the 60 cases: the column label "Best" indicates the number of cases with the same
duration as the current best result given by MA (data from PSPLIB, August 31, 2013); The calculation
130
results in Table 7 show that: when pc=O. Algorithm performance is best when 80. Next, set P. =O.
80, test the effect of pop Size, Gen, and master on MA performance. The calculation results are
shown in Table 8.
PC Sum Best Avg. Dev (%) Avg. CPU Max. CPU
0.75 7391 27 29.94 55.26 104.38

0.81 7384 28 29.84 55.48 102.79
0.86 7388 29 29.92 58.35 120.05
0.91 7393 25 30.01 59.76 114.55
Table 7: Impact of pe Value on MA Execution Results (pop Size=80, Gen=30, mart=20).
pop size Gen Max Iter Sum Best Avg. Dev (%) Avg. CPU Max. CPU
60 31 21 7388 28 29.89 54.21 100.65
60 41 16 7392 28 29.94 56.32 108.49
60 41 21 7388 28 29.87 68.24 137.28
60 51 16 7393 26 29.97 66.79 120.03
60 51 21 7388 27 29.92 80.51 147.53
80 31 16 7384 27 29.84 55.49 102.79
80 31 21 7384 28 29.84 55.48 102.79
80 41 16 7378 27 29.77 74.96 137.48
80 41 21 7379 28 29.72 91.62 173.35
80 51 16 7387 27 29.85 88.95 168.32
80 51 21 7369 28 29.59 109.89 213.26
100 31 16 7385 29 29.82 77.49 140.18
100 31 21 7382 29 29.77 94.15 183.19
100 41 11 7394 27 29.99 76.75 148.84
100 41 16 7376 28 29.69 99.03 194.08
100 41 21 7374 29 29.65 115.49 219.71
100 51 11 7373 27 29.67 91.28 165.65
100 51 16 7372 29 29.63 117.02 214.82
100 51 21 7372 32 29.61 137.11 244.65
Table 8: Impact of the parameters pop size, Gen and mall on MA execution results (p=0.80).
When maxlter=5, none of the parameter combinations satisfy the "Avg.Dev." value of less than
30.00%. In the parameter combinations that satisfy the "Avg.Dev." value of less than 30.00%, the
influence of different combinations on the execution result of MA is not obvious. The random
calculation result shows that the value of "Avg.Dev." is the smallest when popSée=80, Gen=50, and
maxlter=20. Therefore, in the subsequent formal testing process, the algorithm parameters are set
as follows: popSme=80, Gen=50, maxBer=20, P. = 0.80, preGen=10 and elitism=5. However, the
computation time of the MA using LSHYBRID is significantly increased compared to the MA using the
other three local search procedures. In terms of comprehensive solution quality and time efficiency,
the MA comprehensive performance using the LSENINSERT local search process is the best. See
Table 9.
In the past few decades, there have been many heuristic algorithms for solving RCPSP. These
algorithms are often coded in different computer languages and run-on different machine
platforms, so it is difficult to compare their performance.

131
Problem set local search J30 J60 J120

Sum LSSWAP 28344 38559 75444
LSINSERT 28318 38395 74472
LSENINSERT 28317 38378 74405
LSHYBRID 28317 38359 74182
Best LSSWAP 461 379 227
LSINSERT 479 416 265
LSENINSERT 481 429 275
LSHYBRID 481 437 293
Avg. Dev (%) LSSWAP 0.09 11.13 32.93
LSINSERT 0.02 10.64 31.23
LSENINSERT 0.01 10.59 31.12
LSHYBRID 0.01 10.54 30.72
Avg. CPU LSSWAP 0.08 1.63 22.22
LSINSERT 0.05 2.66 21.95
LSENINSERT 0.04 3.14 27.64
LSHYBRID 0.06 10.95 107.13
Max. CPU LSSWAP 1.37 6.87 41.23
LSINSERT 3.74 12.24 42.04
LSENINSERT 4.42 14.58 53.68
LSHYBRID 4.06 50.12 211.27
Table 9: Main Calculation Results of MA.

To have a fair comparison criterion, a method is proposed to ensure fair comparison of all algorithms
on the PSPLIB case set, that is, using generating 1000, 5000 and 50000 solutions as the termination
condition of the algorithm. Tables 10 to 12 show the comparison between the calculation results of
MA using the local search process LSENINSERT and the calculation results of six excellent algorithms
for the case sets J30, J60 and J20. The 6 excellent algorithms selected include the 4 algorithms
presented in the review.
algorithm Maximum number of scheduling plans

1000 5000 50000
Ranjbar, etc. 0.11 0.04 0.00
Kochetov and Stolyar 0.11 0.05 0.00
MA+ LSENINSERT 0.12 0.05 0.00
Debels et al 0.28 0.12 0.02
Debels and Vanhoucke 0.13 0.05 0.03
Valls et al 0.28 0.07 0.03
Alcaraz et al 0.26 0.07 0.04
Table 10: Average deviation between solution result and optimal solution (J30).
It can be seen from Table 10 to Table 12 that the solution time is appropriately increased under
acceptable conditions. For the J30, J60 and J120 case sets, the calculation results of MA are better
than those of the compared algorithms.
132

1000 5000 50000
Ranjbar, etc. 11.58 11.08 10.65

MA+ LSENINSERT 11.72 11.29 10.64
Debels et al 11.74 11.11 10.72
Valls et al 11.57 11.11 10.74
Alcaraz et al 11.89 11.19 10.85
Table 11: Average Deviation between Solution Results and Critical Path Length (J60).

1000 5000 50000
Ranjbar, etc. 35.09 34.12 31.46

MA+ LSENINSERT 35.19 34.12 31.47
Debels et al 35.23 33.11 31.58
Valls et al 11.57 11.11 10.74
Alcaraz et al 11.89 11.19 10.85
Table 12: Average Deviation between Solution Results and Critical Path Length (J20).
5 CONCLUSION
In today's more complex economic development situation, enterprises should correctly analyze and
understand the system problems, strategy formulation stability, personnel allocation problems, and
personnel mechanism problems existing in human resource management activities. This paper uses
MA to encode individuals with task linked list, uses single-point crossover operator as recombination
operator, and applies five local search processes to improve offspring individuals. The algorithm
performance is tested using many cases from the case sets J30, J60 and J120 in PSPLIB. In addition,
we use the model and method of multi-objective optimization problem, and the selected test cases
are the optimization functions of two objectives. At the same time, the non-dominated set sorting
algorithm in IMOCS is not limited by the number of objectives and decision variables. It is feasible
to apply the IMOCS algorithm to the multi-objective optimal management in the actual human
resource project.
Junxin Chang , https://orcid.org/0009-0009-6538-2752
Shuo Wang, https://orcid.org/0009-0000-7109-3097
Quan Li, https://orcid.org/0009-0003-6210-9124
REFERENCES
[1] Arrieta, A. B.; Díaz-Rodríguez, N.; Del Ser, J.; Bennetot, A.; Tabik, S.; Barbado, A.; Herrera,
F.: Explainable Artificial Intelligence (XAI), Concepts, taxonomies, opportunities and challenges
toward responsible AI, Information Fusion, 58, 2020, 82-115.
https://doi.org/10.1016/j.inffus.2019.12.012
[2] Belizón, M. J.; Kieran, S.: Human resources analytics: A legitimacy process, Human Resource

133
Management Journal, 32(3), 2022 , 603-630. https://doi.org/10.1111/1748-8583.12417

[3] Boon, C.; Den Hartog, D. N.; Lepak, D. P.: A systematic review of human resource
management systems and their measurement, Journal of Management, 45(6), 2019 , 2498-
2537. https://doi.org/10.1177/0149206318818718
[4] Boudlaie, H.; Amoozad Mahdiraji, H.; Shamsi, S.; Jafari Sadeghi, V.; Garcia-Pereze, A.:
Designing a human resource scorecard: An empirical stakeholder-based study with a company
culture perspective, Journal of Entrepreneurship, Management and Innovation, 16(4), 2020,
113-147. https://doi.org/10.7341/20201644
[5] Camacho, D.; Panizo-LLedot, Á.; Bello-Orgaz, G.; Gonzalez-Pardo, A.; Cambria, E.: The four
dimensions of social network analysis: An overview of research methods, applications, and
softwaretools, Information Fusion, 63, 2020, 88-120.
[6] Chams, N.; García-Blandón, J.: On the importance of sustainable human resource management
for the adoption of sustainable development goals, Resources, Conservation and
Recycling, 141, 2019, 109-122. https://doi.org/10.1016/j.resconrec.2018.10.006
[7] Diez-Olivan, A.; Del Ser, J.; Galar, D.; Sierra, B.: Data fusion and machine learning for
industrial prognosis: Trends and perspectives towards Industry 4.0, Information Fusion, 50,
2019, 92-111. https://doi.org/10.1016/j.inffus.2018.10.005
[8] Doz, Y.: Fostering strategic agility: How individual executives and human resource practices
contribute, Human Resource Management Review, 30(1), 2020, 100693.
https://doi.org/10.1016/j.hrmr.2019.100693
[9] Evgenievna, Z. I.: Modern aspects of the application of information and communication
technologies in the management of the statistical industry of the Republic of
Uzbekistan, ACADEMICIA: An International Multidisciplinary Research Journal, 9(8), 2019, 59-
69. https://doi.org/10.5958/2249-7137.2019.00092.2
[10] Ghamisi, P.; Rasti, B.; Yokoya, N.; Wang, Q.; Hofle, B.; Bruzzone, L.; Benediktsson, J. A.:
Multisource and multitemporal data fusion in remote sensing: A comprehensive review of the
state of the art, IEEE Geoscience and Remote Sensing Magazine, 7(1), 2019, 6-39.
https://doi.org/10.1109/MGRS.2018.2890023
[11] Hafezalkotob, A.; Hafezalkotob, A.; Liao, H.; Herrera, F.: An overview of MULTIMOORA for
multi-criteria decision-making: Theory, developments, applications, and
challenges, Information Fusion, 51, 2019, 145-177.
[12] Hamadamin, H. H.; Atan, T.: The impact of strategic human resource management practices
on competitive advantage sustainability: The mediation of human capital development and
employee commitment, Sustainability, 11(20), 2019, 5782.
https://doi.org/10.3390/su11205782
[13] Haque, A. B.; Bhushan, B.; Dhiman, G.: Conceptualizing smart city applications:
Requirements, architecture, security issues, and emerging trends, Expert Systems, 39(5),
2022 , e12753. https://doi.org/10.1111/exsy.12753
[14] Harsch, K.; Festing, M.: Dynamic talent management capabilities and organizational agility—
A qualitative exploration, Human Resource Management, 59(1), 2020, 43-61.
https://doi.org/10.1002/hrm.21972
[15] Kumar, D. P.; Amgoth, T.; Annavarapu, C. S. R.: Machine learning algorithms for wireless
sensor networks: A survey, Information Fusion, 49, 2019, 1-25.
[16] Lau, B. P. L.; Marakkalage, S. H.; Zhou, Y.; Hassan, N. U.; Yuen, C.; Zhang, M.; Tan, U. X.:
A survey of data fusion in smart city applications, Information Fusion, 52, 2019, 357-374.
[17] Liu, Y.; Zhang, L.; Yang, Y.; Zhou, L.; Ren, L.; Wang, F.; Deen, M. J.: A novel cloud-based
framework for the elderly healthcare services using digital twin, IEEE Access, 7, 2019, 49088-
134
49101. https://doi.org/10.1109/ACCESS.2019.2909828
[18] Lu, Y.: The blockchain: State-of-the-art and research challenges, Journal of Industrial
Information Integration, 15, 2019, 80-90. https://doi.org/10.1016/j.jii.2019.04.002
[19] Qi, L.; Hu, C.; Zhang, X.; Khosravi, M. R.; Sharma, S.; Pang, S.; Wang, T.: Privacy-aware
data fusion and prediction with spatial-temporal context for smart city industrial
environment, IEEE Transactions on Industrial Informatics, 17(6), 2020, 4159-4167.
https://doi.org/10.1109/TII.2020.3012157
[20] Roscoe, S.; Subramanian, N.; Jabbour, C. J.; Chong, T.: Green human resource management
and the enablers of green organisational culture: Enhancing a firm's environmental
performance for sustainable development, Business Strategy and the Environment, 28(5),
2019, 737-749. https://doi.org/10.1002/bse.2277
[21] Sima, V.; Gheorghe, I. G.; Subić, J.; Nancu, D.: Influences of the industry 4.0 revolution on
the human capital development and consumer behavior: A systematic
review, Sustainability, 12(10), 2020, 4035. https://doi.org/10.3390/su12104035
[22] Tang, S.; Shelden, D. R.; Eastman, C. M.; Pishdad-Bozorgi, P.; Gao, X.: A review of building
information modeling (BIM) and the internet of things (IoT) devices integration: Present status
and future trends, Automation in Construction, 101, 2019, 127-139.
https://doi.org/10.1016/j.autcon.2019.01.020
[23] Yli-Ojanperä, M.; Sierla, S.; Papakonstantinou, N.; Vyatkin, V.: Adapting an agile
manufacturing concept to the reference architecture model industry 4.0: A survey and case
study, Journal of Industrial Information Integration, 15, 2019, 147-160.
https://doi.org/10.1016/j.jii.2018.12.002
[24] Yong, J. Y.; Yusliza, M. Y.; Ramayah, T.; Chiappetta Jabbour, C. J.; Sehnem, S.; Mani, V.
Pathways towards sustainability in manufacturing organizations: Empirical evidence on the role
of green human resource management, Business Strategy and the Environment, 29(1), 2020,
212-228. https://doi.org/10.1002/bse.2359
[25] Zhang, Y. D.; Dong, Z.; Wang, S. H.; Yu, X.; Yao, X.; Zhou, Q.; Gorriz, J. M.: Advances in
multimodal data fusion in neuroimaging: overview, challenges, and novel orientation,
Information Fusion, 64, 2020, 149-187. https://doi.org/10.1016/j.inffus.2020.07.006


Iee Digital Marketing

Uploaded by

Copyright:

Available Formats

Iee Digital Marketing

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Iee Digital Marketing

Uploaded by

Copyright:

Available Formats

119

Digital Marketing Strategies Leveraging Data Fusion and

Junxin Chang1* , Shuo Wang2 and Quan Li3

Corresponding author: Junxin Chang,[email protected]

Abstract. In the new normal economic development environment, strengthening

Keywords: project planning;human resource management;Digital Marketing

Computer-Aided Design & Applications, 21(S4), 2024, 119-134

2.1 Problems Existing in Human Resource Management

Computer-Aided Design & Applications, 21(S4), 2024, 119-134

2.2 The Role of Human Resource Management

Computer-Aided Design & Applications, 21(S4), 2024, 119-134

2.3 Resource Constrained Project Scheduling Problem

3.1 Memetic Algorithm

Whether the stop Output

Cross Ten Local

Figure 1: Memetic Algorithm Flow.

3.2 Solving the MA of RCPSP

3.3 Sorting Of Non-Dominated Sets

4.1 Evaluation Indicators

Computer-Aided Design & Applications, 21(S4), 2024, 119-134

4.2 Non-Dominated Set Sorting Performance Analysis

algorithm SCH ZDT1 ZDT2 ZDT3 ZDT4 LZ

algorithm SCH ZDT1 ZDT2 ZDT3 ZDT4 LZ

Computer-Aided Design & Applications, 21(S4), 2024, 119-134

Figure 2: Pareto Frontier of Test Cases.

Figure 3: GD Convergence Curve of Test Cases.

Computer-Aided Design & Applications, 21(S4), 2024, 119-134

Figure 4: Variation Curve of an in ZDT4.

algorithm SCH ZDT1 ZDT2 ZDT3 LZ

Table 3: GD Values Calculated by IMOCS and MOCS.

algorithm ZDT1 ZDT2 ZDT3 ZDT4

Table 4: GD mean, and variance calculated by IMOCS and NSPSO.

algorithm ZDT1 ZDT2 ZDT3 ZDT4

Table 5: Mean and Variance of Diversity Calculated by IMOCS and NSPSO.

algorithm SCH ZDT1 ZDT2 ZDT3 LZ

Table 6: GD Values Calculated by IMOCS and NSGA.

4.3 MA Performance Analysis

PC Sum Best Avg. Dev (%) Avg. CPU Max. CPU

0.75 7391 27 29.94 55.26 104.38

Computer-Aided Design & Applications, 21(S4), 2024, 119-134

Problem set local search J30 J60 J120

Table 9: Main Calculation Results of MA.

algorithm Maximum number of scheduling plans

algorithm Maximum number of scheduling plans

Ranjbar, etc. 11.58 11.08 10.65

algorithm Maximum number of scheduling plans

Ranjbar, etc. 35.09 34.12 31.46

Computer-Aided Design & Applications, 21(S4), 2024, 119-134

Management Journal, 32(3), 2022 , 603-630. https://doi.org/10.1111/1748-8583.12417

Computer-Aided Design & Applications, 21(S4), 2024, 119-134

You might also like