Graph Convolutional Networks with Knowledge Graph for Myers-Briggs

Type Indicator

Heru Mardiansyah, Saib Suwilo, Erna Budhiarti Nababan and Syahril Efendi

Universitas Sumatera Utara, Medan, Indonesia

Keywords:

Knowledge Graph, Graph Convolutional Networks, Myers-Briggs Type Indicator.

Abstract:

The order of vertices in a graph is very important because the graph is oriented. Or the vertices are not im-

portant because they are not oriented. The graph of data is a heterogeneous polydigraph, called a directed

vertex set, with many edges between any two vertices. Information is created by establishing real-world rela-

tionships between graphics and objects. This study was conducted to improve machine learning performance

by proposing a theoretical model for the data used and creating a graph convolutional network (GCN) model

for training the data. Data are created from a low-dimensional (latent) space, but can only be observed in

a high-dimensional (observation) space. The results of these studies may not always yield the same results

because they were not measured on the same person, were unreliable, or the results obtained did not provide

consistent results. MBTI tests may change at any time. It is obtained according to the result of a person’s

mood. This MBTI method is often considered weak and unscientiﬁc, so it must be tested with 200 iterations

on the GCN. The resulting GCN scores are 89.8% accuracy and 2.78 Test Loss.

1 INTRODUCTION

MBTI (Myers-Briggs Type Indicator) gives a simple

description in a manner psychometric about type per-

sonality. Although characterization is short This is

Possibly useful in some context apply (in predicting

style behavior characteristics individual, intellectual,

and interpersonal), exists limitations to the psycho-

metric clear instrument (Albrecht et al., ). The MBTI

is a system-type personality that divides people into

16 types of different personalities of 4 parts: Introver-

sion (I) - Extroversion (E), Intuition (N) - Feeling (S),

Thinking (T) - Feeling (F), Judging (J) - Perceiving

(P), (You can read more carry on about What mean-

ing here).

For example, someone with more introversion, in-

tuition, thinking, and perception is called an INTP

inside the MBTI system, and there are Lots of

component-based modeling personality or describe

preference or the person’s behavior based on appoint-

ment (Altuner and Kilimci, ).

This is one tests the most popular personality in

the world. This is used For business, online, fun, re-

search, and more. A simple Google search will dis-

close all method different that has been used to test

from time to time. Can be said that testing This Still

very up to date its use worldwide. From the corner

view scientiﬁc or psychological based on Carl Jung’s

work on function cognitive, that is Jungian typology.

This is a model of eight functions, thought processes,

or methods to think differently that has been sug-

gested for is inside the mind. Work This has changed

become some system with different personalities for

facilitating it, the most popular Of course just is the

MBTI (Hogan et al., ). recently, its usage/validity

questioned because, among other things, it doesn’t

can dependable in experiments around. However, it

has still become a very tool useful in many ﬁelds, and

the goals of this data set are to determine if there is a

possible pattern recognized in type and style writing

certain and, more general, for test validity test mo-

ment analysis, predict, or classify behavior

2 METHODOLOGY

The vertices in a graph is very important so that the

graph is directed, or the vertices of the graph are not

important, so they are not directed. The knowledge

graph is a heterogeneous multidigraph which is called

a sequence of directed vertices and has many edges

between the two nodes. A knowledge is created by

making a real relationship between graphics and ob-

jects. The knowledge graph (KG), also known as

Mardiansyah, H., Suwilo, S., Nababan, E. and Efendi, S.

Graph Convolutional Networks with Knowledge Graph for Myers-Briggs Type Indicator.

DOI: 10.5220/0012448500003848

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Advanced Information Scientiﬁc Development (ICAISD 2023), pages 285-288

ISBN: 978-989-758-678-1

285

a database,representation structured from describing

facts and gathering description entity related, related,

and description semantics entity. Difference between

a database and a Knowledge graph with a scheme

other deﬁned as structured, homogeneous, and stable,

making data graph can be scaled. KG’s advantage is a

more good representation of the heterogeneous object

using room integrated to connect it (Liemena et al., ).

At the actual data graph moment, this tends not to

be a complete and necessary inference engine in pre-

dicting the links between entities available in KG and

the complete missing facts. Classiﬁcation connection

or inference from available KG data called Link Div-

ination. An example of what it looks like is shown in

Figure 1 (Rajabi and Etminani, ).

Figure 1: Sample KG Two Nodes.

Study This is done to help increase Machine

Learning performance with propose a theoretical

model for the data used and building a Graph Con-

volutional Network (GCN) model for train data. Data

is created from room dimension low (latent) but only

can observe in dimensional (observational) space high

(Ramezani et al., ). This means that data is real de-

pending on many small factors, however, stretched in

a manner artiﬁcial to appear to depend on a lot of fac-

tors. GCN only see room data observation and cre-

ate prediction more accurately based on the data used.

Data used though simple will but no trivial matter (El-

hammadi and B, ). The GCN model in research This

is as follows:

3 RESULTS AND DISCUSSIONS

This research requires experimental data to be tested

on the GCN method regarding the MBTI so that later

it can be implemented in the MBTI. This is an experi-

mental method using data obtained from Kaggle. The

data has been analyzed so that it has column types to

identify. Analyzed data in which the ﬁnal column is

Figure 2: GCN Models.

considered the target, and the other columns are the

attributes. The shared datasets become training sets,

validation sets, and test sets.

3.1 Testing

Data has been analyzed to have a type column for

identiﬁcation. Moment analyzes data, column ﬁnal

is treated as target and column other will be enforced

as input ﬁeld. Shared dataset into training data, vali-

dation sets, and testing data (Wang et al., ). The fol-

lowing is the data table used.

The dataset used in the work consists of tagged

tweets. One of 16 MBTI types. These tags are a com-

bination of four letters. Each character matches. The

ﬁrst or second character of four MBTI class character-

istics. The dataset consists of 8660 rows. Distribution

of MBTIs. The characteristics of each class (8600

lines) are as follows.

a. Introversion (I ): 6664

b. Extroversion (E): 1996

c. Recognition:7466

d. Intuition (N): 1194

e. Think (T): 4685

f. Feeling (F) : 3975

g. Judging (J): 5231

h. Perceiving (F): 3429.

GCN has data representation of chart usually use

adjacency matrix and use the feature as input. Fea-

ture This is used as a property of the nodes on the

graph and presented become numbers. Feature This

made example of many groups that have social net-

works that become rejected measuring patience from

some people. Matrix feature This is a 2-dimensional

shape. On-line matrix feature i.e. a vector of similar

features with knot graph, from facet size on each knot.

On the column, the matrix will be the same with fea-

tures certain can understand nodes and their graphs.

ICAISD 2023 - International Conference on Advanced Information Scientiﬁc Development

286

For example, in research this does testing against the

Myers-Briggs Indicator personality namely:

a. Extraversion vs. Introversion

b. Feel vs. Intuition

c. Thinking vs. Feel

d. Assess vs. Perceive

3.2 Test Results

This research requires experimental data to be tested

on the GCN regarding the Myers-Briggs Type Indi-

cator so that later it can be implemented in users of

social networks. This is an experimental method us-

ing data obtained from Kaggle (Yani and Krisnadhi,

). So that it can be simulated to see the achievement

of searching for the highest accuracy and can be used

in a psychological test that is designed to measure a

person’s purely psychological basic preferences. The

resulting achievement was successful in obtaining the

highest value from the epoch experiment carried out

in the analysis process. The table below will be shown

the results of data analysis from epochs 1 to 200.

Comparison of results Among target networks (0 and

1). More clear information could be seen in the table

under this:

Figure 3: GCN Process.

Figure 4: Epoch Test.

This research was conducted to design a person-

ality test myers-briggs type indicator (mbti) using the

knowledge graph method and predict personality.

4 CONCLUSIONS

In the results study, This No can produce always re-

sults same, because No measure from One the same

Figure 5: KG of Myers-Briggs Type Indicator.

individual, and no have reliability or the resulting re-

sults No get consistent results. The MBTI test is also

available depending on the results atmosphere heart

someone who can capricious any time. the MBTI is a

frequent method considered weak and not scientiﬁc,

then need exists testing to use GCN with iterations as

many as 200. Results obtained with the GCN evalu-

ation are 89% accuracy produced and the test loss is

2.78. Result of knowledge graph in p reference third

focus in study How method in making the decision.

Someone decides in a manner objective or based on a

hunch, a (T) if decide in a manner objective, and an

(F) if weigh everything with consider circumstances

personal. Because someone’s objective and biased

preferences logic (T) is not mean they have no feel-

ings. only Because somebody’s own preference this

(F) feeling No means they No think about something.

The last preference is determination style alive, this

will evaluate according to the information data ﬁlled

out, for the group becomes someone who plans or is

ﬂexible.

REFERENCES

Albrecht, J., Belger2, A., Blum2, R., and Zimmermann, R.

Business analytics on knowledge graphs for market

trend analysis *.

Altuner, A. and Kilimci, Z. A novel deep reinforcement

learning based stock price prediction using knowledge

graph and community aware sentiments. Turkish Jour-

nal of Electrical Engineering and Computer Sciences,

30(4):1506–1524.

Elhammadi, S. and B. Financial knowledge graph construc-

tion.

Hogan, A., Blomqvist, E., Cochez, M., d’Amato, C., Melo,

G., Gutierrez, C., Gayo, J., Kirrane, S., Neumaier, S.,

Polleres, A., Navigli, R., Ngomo, A.-C., Rashid, S.,

Rula, A., Schmelzeisen, L., Sequeda, J., Staab, S., and

Zimmermann, A. Knowledge graphs.

Liemena, F., Novianus Palit, H., and Tjondrowiguno,

A. Evaluasi kinerja penggabungan knowledge graph

embedded-based question answering dan transp pada

data freebase.

Rajabi, E. and Etminani, K. Knowledge-graph-based ex-

plainable ai: A systematic review. Journal of Infor-

mation Science.

Graph Convolutional Networks with Knowledge Graph for Myers-Briggs Type Indicator

287

Ramezani, M., Feizi-Derakhshi, M., and Balafar, M.

Knowledge graph-enabled text-based automatic per-

sonality prediction. Computational Intelligence and

Neuroscience.

Wang, X., Chen, L., Ban, T., Usman, M., Guan, Y., Liu,

S., Wu, T., and Chen, H. Knowledge graph qual-

ity control: A survey. Fundamental Research, 1(ue

5):607–626.

Yani, M. and Krisnadhi, A. Challenges, techniques, and

trends of simple knowledge graph question answer-

ing: A survey. In Information (Switzerland. Vol. 12,

Issue 7). MDPI AG.

ICAISD 2023 - International Conference on Advanced Information Scientiﬁc Development

288