An Information Sharing Method for Skilled Management Operations

based on Bayesian Network Inference

Takayuki Kataoka

, Kazumoto Tanaka

, Masakazu Kanezashi

and Makoto Hasegawa

Faculty of Engineering, Kinki University, Takaya Umenobe 1, Higashi-Hiroshima, Hiroshima, Japan

LORIA, UMR7503, 54506, Vandoeuvre-les-Nancy, France

Keywords: Bayesian Network, Information Sharing, Management Operation.

Abstract: Given the poor state of the economies all over the world, almost every manufacturing site has been

supported by a lot of part-time, temporary, or mid-career personnel. And expert managers of front-line

workers must design more complex human resource strategies that take into consideration the workers’

skills. However, tacit knowledge existing only in the minds of expert managers is very difficult to capture

with most organizations depending entirely on the explicit knowledge. Therefore, the purpose of our study

is to develop a model with a bayesian network using the operation histories of expert managers, and to

verify some factors that would make it easier for nonexperts to assign human resources. First, the operation

histories are collected. Next, some differences of human resource planning procedures for expert managers

and nonexperts are discussed by dividing into the purposes of either minimizing makespan or workload.

Finally, the effectiveness of the expert managers’ operations is verified by constructing a bayesian network

model based on the operation histories, and is discussed by way of probabilistic inference.

1 INTRODUCTION

Recently, the number of regular, full-time

employees has been decreasing at general

production factories. On the other hand, the number

of part-timers or temporary employees has been

increasing. For a long time, most companies had

recruited, trained and held onto many talented

people as their regular employees. However, it is

difficult to support all of the regular employees

without outsourcing due to poor economic

conditions. Therefore, the leaders of the front-line

workers must assign each worker appropriate tasks

according to the quantity and quality of work.

Mohanty et al. discussed the evolution of a

decision support system (DSS) for human resource

(HR) planning in a petroleum company. The DSS

helped the HR division of the company analyze HR

decisions to overcome many problems, to cut down

on delays in implementing new projects and to

expand the business into new areas (Mohanty,

1997). Further, Parush et al. showed the impact of

visualization and contextual factors on performance

with enterprise resource planning (ERP) systems.

They showed that some graphic information

visualization displays for ERP systems can increase

the probability of a successful implementation and

enhance the capabilities of the human operators

(Parush, 2007).

Furthermore, Abdinnour-Helm et al. studied the

pre-implementation attitudes and organizational

readiness for implementing an ERP system. Despite

an extensive amount of time, money and effort, the

length of time with a firm and position had a greater

impact on attitudes toward ERP capabilities, value,

acceptance and timing than high levels of pre-

implementation involvement (Abdinnour-Helm,

2003). Youngberg et al. discussed the determinants

of professionally autonomous end user acceptance in

an ERP system environment. The study surveyed 66

professionally autonomous end users and gathered

information on their perceptions related to several

technology acceptance factors for a newly installed

ERP system component (Youngberg, 2009).

On the other hand, Corominas et al. studied the

planning of annualized hours with a finite set of

weekly working hours and cross-trained workers

(Corominas, 2007). Lusa et al. also attempted to

determine the most appropriate set of weekly

working hours for planning annualized working

time. Their paper proposes a method for selecting

the most appropriate set of weekly working hours

257

Kataoka T., Tanaka K., Kanezashi M. and Hasegawa M..

An Information Sharing Method for Skilled Management Operations based on Bayesian Network Inference.

DOI: 10.5220/0004116102570260

In Proceedings of the International Conference on Knowledge Management and Information Sharing (KMIS-2012), pages 257-260

ISBN: 978-989-8565-31-0

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

and establishing an annual plan or working time for

each worker as a way of optimizing service levels

(Lusa, 2008).

However, most HR planning systems do not

consider workers’ skills. According to hearing

surveys with some small and medium-sized

enterprises (SMEs), it is enough to have even the

simple support tool that a sub-leader, rather than the

leader, can redesign the current strategy using an

interactive interface.

Thus, it is used that a tool which has already

developed with “An Inference Method of

Management Operations using Bayesian Networks

(Kataoka et al, 2010)” using the Program Evaluation

and Review Technique (PERT) to collect a lot of

operation histories. That is the reason why our

cooperative researcher is a middle iron work’s

company in Japan.

Furthermore, our paper has also already

suggested an inference method of management

operations using bayesian networks (Kataoka et al.,

2011). However, it is not enough that the

effectiveness of an inference method is shown in

only a sample data of human resource planning.

Therefore, this paper researches to develop a

model with a bayesian network using the operation

histories of expert managers, and to verify some

factors that would make it easier for nonexperts to

assign human resources.

2 BAYESIAN NETWORK

A Bayesian network is one of the probabilistic

models which express the dependence between

random variables by a conditional probability.

Figure 1: Bayesian network model.

Based on a sample data, each variable has

already classified into three categories and each

causal relationship has also already specified in our

paper as shown in Figure.1 (Kataoka et al., 2011).

Firstly, some factors like worker’s skill or

operational targets are arranged as parent nodes.

Next, operational processes from 1st operation to

10th are arranged with causal relationships to the

increase or decrease value of makespan and

workload. Every operational process is recorded to

show the difference between the expert manager and

nonexperts clearly. Lastly, some results are arranged

with causal relationships to operational processes.

3 SKILLED OPERATIONS

3.1 Makespan and Workload

In the operation purpose to minimize makespan, it is

verified that the result of the expert manager who

will greatly decreases makespan time at the first

stage is better. Figure 2 shows a graph where the

mean value of the increase or decrease in workload

when makespan times of each operation decrease by

one unit is shown. There are not so many differences

in the increase or decrease of both experts and

nonexperts at the first stage. However, the expert

manager greatly decreases the workload in the latter

half. As a result, it can be judged that the expert

manager is finally obtaining a great result for both

makespan and workload.

Figure 2: The increase or decrease ratio of reworks to a

decrease per unit time.

3.2 Reworks

Figure 3 shows the mean value of makespan and

workload in the case that some reworks will occur or

not to minimize workload. The decision to minimize

Results

Factors

T : Makespan

W: Workloa

Operator Criteria

1stT

1stW

2ndT

2ndW

3rdT

3rdW

4thT

4thW

5thT

5thW

6thT

6thW

7thT

7thW

8thT

8thW

9thT

9thW

10thT

10thW

Total W

Last T

WConst

Back

T : Makespan

W: Workload

Operational Process

KMIS2012-InternationalConferenceonKnowledgeManagementandInformationSharing

258

workload is done without decreasing makespan.

Therefore, workload is approached to zero at the last

stage, and makespan is shortened. As a result, many

reworks will occur because operators fall greatly

into disorder.

Figure 3: Difference of procedure in reworks.

4 EXPERIMENT

4.1 Sample Data

As a comprehensive experiment, 4 sample data is

prepared as shown in Table 1. Some small scale

projects might be defined to have 100-200 hours as

total operation time (TepStep, 2011).

Therefore our study assumes that the number of

maximum process is 35 and the number of minimum

process is 14 in case of 7 hours per process.

Table 1: Sample data for comprehensive experiment.

4.2 Experimental Results

Modes of experimental results for each sample are

shown in Table 2. CPT means Conditional

Probability Table. In case of many processes

(Sample 0 and 1), experts could operate the tool

better than nonexperts. Especially, it seems that

nonexperts could not understand the way to decrease

total workload as shown in the results of ‘Minimum

Total Workload’.

In case of less workers (Sample 1 and 3), it

seems that nonexperts tend to rework a lot.

Table 2: The first result of comprehensive experiment.

4.3 Instructions to Nonexperts

Instructions expected to be effective for the

improvement are shown in the following.

Operate to decrease the schedule from the left or

the right point to avoid overlaps on critical path.

Operate up to 1st-4th operation (Sample0:1,

Sample1:2, Sample2: 4, Sample3: 3) to minimize

makespan as a first object.

Assign a worker who takes a long operation time

for a task in case of ‘Minimum Total Workload’.

Operate to decrease the most overlapped

workload for a task in case of ‘Minimum Total

Workload’.

4.4 Verified Results

The result that nonexperts try it again based on the

instructions of 4.3 is shown in Table 3.

As a result, almost samples are improved.

Additionally, nonexperts got to tend not to rework as

shown in Table 4.

4.5 Inductive Modelling

An information sharing model (:inductive

modelling) based on our experimental results is

shown in Figure 4.

Firstly, a set of experts’ skilled operation

histories is prepared. Next, nonexpert’s operations

are compared with a set of skilled operations in real

time, and some instructions are given to a nonexpert

based on differences and causal relationships of

bayesian networks. Lastly, a bayesian database is

Process Worker

Sample0 35 10

Sample1 35 4

Sample2 14 10

Sample3 14 4

Mode CPT Mode CPT Mode CPT Mode CPT

Sample0 Makespan 136~140 0.72 136~140 0.30 216~220 0.67 206~210 0.30

Workload 131~135 0.25 136~140 0.08 21~25 0.24 46~50 0.07

Sample1 Makespan 131~135 0.68 141~145 0.25 206~210 0.29 216~220 0.14

Workload 176~180 0.08 196~200 0.04 91~95 0.21 151~155 0.05

Sample2 Makespan 66~70 0.69 66~70 0.53 66~70 0.50 66~70 0.38

Workload 0 0.69 0 0.35 0 0.08 0 0.73

Sample3 Makespan 96~100 0.90 96~100 0.90 111~115 0.53 111~115 0.53

Workload 11~15 0.92 11~15 0.92 1~5 0.71 1~5 0.71

Minimum Makespan

Expert Nonexpert Expert Nonexpert

Minimum Tota l Workload

T : Makespan

W: Workloa

AnInformationSharingMethodforSkilledManagementOperationsbasedonBayesianNetworkInference

259

updated, and next operation is considered with an

object.

Table 3: The result of comprehensive experiment.

Table 4: The total number of reworks.

Figure 4: Information sharing model.

5 CONCLUSIONS

This paper showed an information sharing model

with a bayesian network using the operation

histories of expert managers, and verified some

factors that would make it easier for nonexperts to

assign human resources.

In the future, the decision of HRP to assume all

situations needs to be verified, and the improvement

of further inference accuracy is requested.

ACKNOWLEDGEMENTS

This work was supported by KAKENHI 23710186.

REFERENCES

Mohanty, R. P. and Deshmukh, S. G., 1997, Evolution of a

decision support system for human resource planning

in a petroleum company, International Journal of

Production Economics, 51, Issue 3, 251-261.

Parush, A., Hod, A. and Shtub A., 2007, Impact of

visualization type and contextual factors on

performance with enterprise resource planning

systems, Computers & Industrial Engineering, 52,

Issue 1, 133-142.

Abdinnour-Helm, S., Lengnick-Hall, M. L. and Lengnick-

Hall, C. A., 2003, Pre-implementation attitudes and

organizational readiness for implementing an

Enterprise Resource Planning system, European

Journal of Operational Research, 146, Issue 2, 258-

273.

Youngberg, E., Olsen, D. and Hauser, K., 2009,

Determinants of professionally autonomous end user

acceptance in an enterprise resource planning system

environment, International Journal of Information

Management, 29, Issue 2, 138-144.

Corominas, A., Lusa, A. and Pastor, R., 2007, Planning

annualized hours with a finite set of weekly working

hours and cross-trained workers, European Journal of

Operational Research, 176, Issue 1, 230-239.

Lusa, A., Pastor, R. and Corominas, A., 2008,

Determining the most appropriate set of weekly

working hours for planning annualized working time,

International Journal of Production Economics, 111,

Issue 2, 697-706.

Kataoka, T., Tanaka, K., and Hasegawa, M., 2010,

Interactive Model for Human Resource Planning In

Operating a group of Different Cycle Time,

Proceedings of The 21st IASTED International

Conference -MODELLING AND SIMULATION-, July

15-17, Banff, Canada, pp.322-329.

Kataoka, T., Kanezashi, M., Morikawa, K. and Takahashi,

K., 2011, An Inference Method of Management

Operations using Bayesian Networks, Proceedings of

21st International Conference on Production Research,

July 31-August 4, Stuttgart, Germany, p.230(6P).

TepStep Japan, 2011, Definition of Operation - Decision

of Project Scale, p5/13.

Expert

Before After

Makespan 146.4 141.8 137.6 3.1%

▲

Total Workload 153.4 152.4 140.3 0.6%

▲

Makespan 205.4 214.8 216.9 4.5%

▽

Total Workload 79.9 36.2 32.3 54.6%

▲

Makespan 148.2 131.9 132.8 10.9%

▲

Total Workload 203.3 161.9 176.1 20.3%

▲

Makespan 210.6 170.6 196.1 18.9%

▲

Total Workload 163.8 120.4 100.7 26.4%

▲

Makespan 67.4 66.0 66.0 2.0%

▲

Total Workload 3.6 0.0 0.0 100.0%

▲

Makespan 88.1 68.1 66.7 22.7%

▲

Total Workload 0.3 0.0 0.2 100.0%

▲

Makespan 92.3 97.2 96.0 5.3%

▽

Total Workload 46.2 15.1 14.5 67.3%

▲

Makespan 129.4 113.2 112.3 12.5%

▲

Total Workload 19.9 6.7 5.5 66.4%

▲

Sample 3

Minimum

Makespan

Minimum

Total Workload

Sample 1

Minimum

Makespan

Minimum

Total Workload

Sample 2

Minimum

Makespan

Minimum

Total Workload

Nonexpert

Improvement

Rate

Sample 0

Minimum

Makespan

Minimum

Total Workload

Before After

Sample0 050

Sample1 0 31 0

Sample2 040

Sample3 3 33 0

Expert

Nonexpert

KMIS2012-InternationalConferenceonKnowledgeManagementandInformationSharing

260