Sentiment Analysis using Sentiwordnet and Machine Learning Approach (Indonesia general election opinion from the twitter content)

Full text

Sentiment Analysis using Sentiwordnet and
Machine Learning Approach
(Indonesia general election opinion from the twitter
content)
Eka Miranda
Information Systems Department,
School of Information Systems
Bina Nusantara University
Jakarta, Indonesia 11480
ekamiranda@binus.ac.id
Mediana Aryuni
Information Systems Department,
School of Information Systems
Bina Nusantara University
Jakarta, Indonesia 11480
mediana.aryuni@binus.ac.id
Edwin Satya Surya
Information Systems Department,
School of Information Systems
Bina Nusantara University
Jakarta, Indonesia 11480
edwin.surya@binus.ac.id
Ricky Hariyanto
Information Systems Department,
School of Information Systems
Bina Nusantara University
Jakarta, Indonesia 11480
ricky.hariyanto@binus.ac.id
Abstract— The computational process of identifying and
categorizing opinions that are expressed in the piece of text
could be employed to determine information insight and the
writer's opinion toward a particular topic. Most sentiment
analysis employed for English text. Contrarily, a plethora
method for sentiment analysis has been reported that the task
stayed an interesting question for Indonesian text. The
invention of machine learning models and broad accessibility
of Twitter data on previous years have derived many
researchers to take a machine learning model to resolve the
sentiment analysis problem. The objective of this study is to
build a sentiment analysis model using Sentiwordnet and
machine learning for Indonesia general election opinion in
Indonesian text from the twitter content. The data of the tweet
was taken namely, the username, and the tweet itself. The
theme of the tweet was the topic related to the 2019 general
election figures, namely Joko Widodo and Prabowo Subianto.
The period of data collection was November 13, 2018, to
January 11, 2019, during the campaign period. The tweet was
in Indonesia language. The result revealed sentiment analysis
with the Naïve Bayes classification method showed 74.94%
accuracy for Joko Widodo topic, while 71.37% accuracy for
Prabowo topic.
Keywords—sentiment analysis, SentiWordNet, twitter, naïve
bayes classification
I. I
NTRODUCTION
In the modern era, communication and socialization
could be done by indirect communication through new media
namely social media. Social media or often called social
network is a tool that could be used to communicate with
each other without direct interaction between individuals [1].
About 120 million Indonesia people use mobile devices,
such as smartphones or tablets to access social media (with
the penetration rate of 45%). Online activity through social
media reaches 37% from the population of in Indonesia in a
week [2]. Over the past 2 years, 90% of all data over the
world was produced from the internet [3]. In one day, the
data created from around the world increased by 2.5
quintillion bytes. The number was produced from the
combination of many tools, including search engines, such as
Google, Yahoo; electronic mail; digital photos; internet of
things, and social media [4]. Facebook produced 6 billion
bytes of data per day. Twitter was successfully produced 500
million bytes of data per day [4]. The computational process
of identifying and categorizing opinions that are expressed in
the piece of text could be employed to determine information
insight and the writer's opinion toward a particular topic [5].
Sentiment analysis is one of the most appealing research
topics in computer science [6]. Sentiment analysis or opinion
mining is one of the main tasks of NLP (Natural Language
Processing) [7].
Zhaou suggested a new method combining social context
and topic context to analyze micro blog sentiment [8]. Fang
presented the sentiment polarity categorization process to
analysis product review data [7]. Naiknaware suggested
social media sentiment analysis using machine learning
classifiers from the online posts from twitter [9]. Hausler
proposed the relation among news-based sentiment, captured
over a machine learning approach, and the US securitized
and direct commercial real estate market [10].
Social media has a great impact on various issues in the
community, in Indonesian people as well. The Indonesian
Presidential Election 2019-2024 issue is the most appealing
for Indonesian people today. Prediction about the winner of
the Indonesia presidential election became a hot and
attractive discussion among Indonesian citizen on social
media as well. The candidate team (Joko Widodo and
Prabowo) have exploited information from various media
(social media as well) to get an insight of citizen opinion
about the candidate. A collective opinion was produced by
social media (Twitter) directly influence to the brand or
public-figure image. Brand image or public-figure image is
the one important issue related to their/him/her credibility.
The positive sentiment to the brand or public-figure produces
a positive impact on the person or the brand and vice versa
[2].
The research employed by the Indonesia Press council
and the Danish embassy analyzed the twitter account profile
and the corresponding between person tendency to the
presidential candidate and the tweet and re-tweet content
[11]. The citizen sentiment about Indonesia president
candidate could be considered as an input for the campaign
strategy. The important impact of social media was
considered by politician and the candidate. Internet and
social media could be a potential tool for the campaign.
978-1-7281-3333-1/19/$31.00 ©2019 IEEE
2019 International Conference on Information Management and Technology (ICIMTech)
62
19-20 Au
g
ust 2019, Jakarta & Bali, Indonesia

Social media and internet could be used as a tool to deliver
and communicate the candidate working plan to the wide
community [12]. Since rich data available on social media
(Twitter) about Indonesia president candidate, so the data
could be used to predict citizen sentiment about the
candidate. One method to analyze the text data is sentiment
analysis.
Despite many studies have proposed sentiment analysis
for text content analysis, to the best of our knowledge, it is
just a few of them has been used the Indonesian text
especially for Indonesia general election (presidential
election) topic. Most sentiment analysis employed for
English text. Hence, the objective of this study is to build a
sentiment analysis model using Sentiwordnet and machine
learning for Indonesia general election (presidential election)
topic in Indonesian text from the twitter content. Sentiment
analysis model could be used as an alternative text analysis
tools instead of a survey approach.
II.
LITERATURE
S
TUDY
A. Sentiment Analysis
Opinion mining (sometimes recognized as sentiment
analysis) referring to the use of natural language processing
(NLP), text analysis and computational linguistics to
identify, extract, quantify, and analyze the content of text
information. Sentiment analysis was broadly used to the
analyze of the customer opinion, for instance, reviews and
survey responses from social media [13].
Sentiment classification is a process of detecting
expression of the text, positive expression or negative
expression about an issue or topic. Sentiment classification
has become a popular technology on Twitter content
analysis. The tweet sentiment classification has been widely
applied in various fields namely politics, social, market share
research, and others [14].
The sentiment analysis process uses the Natural
Language Processing (stemming, part-of-speech tagging) and
also uses the additional resources (thesaurus, a dictionary of
sentiments or emotions) [15]. Type of sentiment analysis
tasks namely opinion mining and emotion mining.
Taxonomy of sentiment analysis tasks is shown in Figure 1.
This study would use the opining mining task.
Fig. 1. Taxonomy of sentiment analysis tasks [16]
B. Classification Type
Sentiment analysis can be done with two approaches,
namely the Lexicon-based and Machine Learning
approaches.
a. Machine leaning approach
The machine learning approach can be grouped
into two main types: supervised and unsupervised.
The success of the two is primarily based on the
selection and extraction of a suitable set of features
used to identify the sentiment [17].
b. The Lexicon-based approach mainly depends on
the sentiment dictionary (a set of terms that are
identified compiled and developed for traditional
communication genres, for example, Opinion
Finder lexicon or even more complex structures for
example ontology, or dictionaries that measure the
semantic orientation of words or phrases).
The sentiment classification type is shown in Figure 2.
This study used machine learning
approach.
Fig. 2. The sentiment classification type [18]
C. Term Weighting
Term weighting is the process of calculating the weights
for each term in the vocabulary. There are two types of the
word weighting scheme namely supervised and unsupervised
word weighting. This study was used unsupervised word
weighting. Unsupervised word weighting is a weighting
scheme that does not depend on the data category. This
method namely: Term Frequency, Inverse Document
Frequency, Term Frequency-Inverse Document Frequency
and Binary. tf–idf (term frequency-inverse document
frequency), is a numerical statistic that is proposed to show
the importance of a word in a document [19]. tf-idf formula
is shown in equation 1 and 2.
W(d, t) = T F(d, t) (1)
d is the document and t is the word. N is the number of
documents, and d
t
is a document that contains the word t.
D. SentiWordNet
SentiWordNet (version 1.0) is a lexical resource where
every word on Wordnet (version 2.0) related to three
numerical scores, namely Obj (s), Post (s) and Neg (s).
978-1-7281-3333-1/19/$31.00 ©2019 IEEE
2019 International Conference on Information Management and Technology (ICIMTech)
63
19-20 Au
g
ust 2019, Jakarta & Bali, Indonesia

These scores defining the objectivity, positivity, and
negativity level in the term that contained in the words.
Each score value ranges from 0.0 to 1.0, and their sum is 1.0
for each synset (synset is a set of one or more synonyms).
SentiWordNet used a random walk algorithm. The theory
for the random walk stage namely, if 2 synset or words have
the same context, they tend to have the same sentiment.
Each synset will be associated with a positive or negative
context. The more relationships with a positive context, the
greater the positive value, and vice versa
[20].
E. Naïve Bayes Classifier
Naive Bayes classifier (NBC) is one of the machine learning
methods that utilizes probability and statistical calculations
to predict probabilities in the future based on previous
experience. This research used the Naive Bayes classifier
due to this method needs a small amount of training data to
appraise the importance of a parameter for classification
[21]. Naive Bayes is a conditional probability-based model
for classification. The problem instance that was given in
this model represented by a vector x = (x
1
…. x
n
) shows
some n features (independent variables) and assigns to the
instance probabilities for every k possible
outcomes or classes C
k
. Bayes' theorem calculated the
conditional probability using the formula in equation 3. [21].
III.
DATA AND METHODS
A. Dataset
The tweet data was taken from Twitter using technology
provided by Twitter for anyone who wants to retrieve their
data for free. Additional tools were made for this purpose,
namely: (1) Twitter Application Programming Interface
(API) Credentials, (2) Python script with the Tweepy library
that will pull data and put it in a CSV file with detailed data
(the date tweet was taken, the user's username, and the tweet
itself). The theme of the tweet was the topic related to the
2019 general election figures, namely Joko Widodo and
Prabowo Subianto. The period of data collection was
November 13, 2018, to January 11, 2019, during the
campaign period. The tweet was in Indonesia language.
B. Methods
This study consists of four main steps, namely Data pre-
processing, determine class label with SentiWordNet,
Sentiment analysis classification and Evaluate the result.
1. Data pre-processing
This step consists of:
a. Tweet selection
This step would remove a tweet that contains
hashtag (#), re-tweet features, duplicating tweets or
links solely.
b. Normalize the sentences
This step would revise word spelling and improve
the word based on the Indonesia dictionary
c. Translate tweet from Indonesia language into
English
d. Data cleansing
This step would remove special characters
(hashtags, periods, commas), usernames, URLs,
HTML tags, words consisting of one character,
excess spaces and number
e. Tokenization and case conversion
This step would parse the sentences, paragraphs, or
documents into smaller parts, called tokens or
independent words. Subsequently, all token would
be converted into the lower case to eliminate the
difference between lower case and capital case.
f. Remove the not alphabet token
2. Determine class label with SentiWordNet
This step would receive every tweet, calculate the value
of each word on the tweet based on SentiWordNet, and
combine each value to find out the positive or negative
value of the word
3. Sentiment analysis classification
a. Term weighting
The TF-IDF method would process the
vectorization of each tweet in the form of text and
change them into numbers.
b. Feature selection
Bigam (part of N-gram) was used as the feature
selection method (word selection). Bigram grouped
words on tweets into groups of words consisting of
2 words to achieve more accurate sentiment
analysis results. Subsequently, k-fold cross-
validation was employed to determine the training
data and test data. 10-fold cross-validation was used
in this study. This is a common technique in k-fold
cross-validation. 9-fold/9 folds was used for
training classifier and 1-fold was used for testing
part.
c. Sentiment analysis classification
Naïve Bayes method was employed for this
classification.
4. Evaluate the result
Evaluation based on confusion matrix was employed to
evaluate the result.
Research framework is shown in Figure 3.
Fig. 3. Research framework
978-1-7281-3333-1/19/$31.00 ©2019 IEEE
2019 International Conference on Information Management and Technology (ICIMTech)
64
19-20 Au
g
ust 2019, Jakarta & Bali, Indonesia

IV.
RESULT AND DISCUSSION
A. Data pre-processing
The words cleansing and normalization process were
performed by removing the duplicate tweet and removing a
tweet that contains the username, emoticon or hashtag solely.
The cleansing and normalization process produced 10662
tweets (6152 tweets for Joko Widodo topic and 6152 tweets
for Prabowo topic respectively). Since the lack of Python
Google Translate capability to translate Indonesian into the
English language subsequently, all words as a result from
cleansing and normalization process were translated into
English through Google translate and Cloud Translation API
from Google Cloud Platform. Data cleansing itself was
performed by removing username, URL, HTML or tags,
eliminating special characters, removing all words that
contain one character, eliminating excess space, tokenizing
and case conversion (convert tweet into lower case),
removing stop-words and lemmatization and POS Tagging.
The lemmatization performed by calling the
WordNetLemmatized function. Subsequently, Part-of-speech
tagging performed to identify words on the sentences. The
word identified as a verb, adjective, adverb and noun. The
script for lemmatization and POS Tagging is shown in
Figure 4.
Fig. 4. Lemmatization and POS Tagging script
B. Determine class label with SentiWordNet
There are two stages to determine class labels namely:
Split the text and Lemmatization the split text. Split the text,
this process would parse the tweets into words using space
characters as references. Lemmatized word, this process aims
to make sure the split text results could back to the base
words without additional letters (the suffix). Subsequently,
score analysis program that designed in this study would
accumulate sentiment score and recognize the positive,
neutral, and negative scores based on SentiWordNet
dictionary for each word from a tweet sent by Twitter users.
Script to accumulate the total score of positive, negative and
neutral sentiment is shown in Figure 5-7 respectively.
Fig. 5. Script to accumulate positive score
If accumulated of positive score > accumulated of
negative score, a tweet would be recognized as positive
sentiment, if accumulated of negative score > accumulated of
positive score, a tweet would be recognized as negative
sentiment, if accumulated of positive score = accumulated of
negative score, a tweet would be recognized as a neutral
score. Neutral score was accumulated from the formula in
equation 3.
Neutral score =
1 – (Positive score + Negative score) (3)
Fig. 6. Script to accumulate negative score
978-1-7281-3333-1/19/$31.00 ©2019 IEEE
2019 International Conference on Information Management and Technology (ICIMTech)
65
19-20 Au
g
ust 2019, Jakarta & Bali, Indonesia

Fig. 7. Script to accumulate neutral score
Table 1 shows the classification result for Joko Widodo
topic and Prabowo topic. Data pre-processing produced
10661 tweet and 6152 tweets for each topic respectively.
69.54% for Joko Widodo topic has a positive score, 10.64%
has neutral score and 19.82% has a negative score. While
61.52% for Prabowo topic has a positive score, 10.71% have
neutral score and 25% has a negative score. This
accumulated score would be used as a reference to define the
sentiment of a tweet.
TABLE I. T
HE
C
LASSIFICATION
R
ESULT
Subject/ Sentiment Positive Neutral Negative
Prabowo Subianto 3955 659 1538
Joko Widodo 7414 1134 2113
C. Sentiment Analysis Classification
This step began with the term weighting process. This
process has used the functions of the Sklearn library to get
the TfidVectorizer function. This function has two
parameters namely: (1) the stop_words parameter to remove
stop-words and (2) ngram_range parameter to select the
word range in the n_gram process. Subsequently, the data
were divided into training data and testing dataset. The
KFold function and 10_split parameter were used to produce
9 fold from the dataset as a training data and 1 fold from the
dataset as testing data. Afterwards, the sentiment analysis
classification was performed using the naïve Bayes
classification method. Each topic (Joko Widodo and
Prabowo) of tweet data would be trained using the Naïve
Bayes classifier. Once the Naïve Bayes classifier has been
trained, the sentiment label prediction would be appended
into each tweet and the data would be stored in an array.
Finally, the actual sentiment label results and the prediction
label results would be used to calculate the accuracy of
Naïve Bayes classifier based on the confusion matrix.
Confusion matrix was used to calculate the accuracy ratio
by dividing correct predictions with total predictions made.
Recall ratio shows part of actual positives was identified
correctly, Precision shows part of positive identifications
was actually correct and F1 Score calculates a mean of
precision and recall. The confusion matrix for the
classification result is shown in Table II and Table III.
TABLE II.
THE CONFUSION MATRIX
:
J
OKO WIDODO
T
OPIC
Positive Negative Neutral
Positive 733.50 7.50 0.40
Negative 159.30 51.80 0.20
Neutral 97.10 2. 70 13.60
Predicted Class
Actual Class
TABLE III.
THE CONFUSION MATRIX
:
P
RABOWO TOPIC
Positive Negative Neutral
Positive 385.20 9.00 1. 30
Negative 115.40 37. 00 1.40
Neutral 450.50 3. 50 16.90
Predicted Class
Actual Class
Sentiment analysis with the Naïve Bayes classification
method showed 74.94% accuracy, 45.23% the average recall
value, 84.61% the average precision value, and 46.62% the
average F1 score for Joko Widodo topic. While 71.37%
accuracy, 49.02% the average recall value, 77.7% the
average precision value, and 50.58% the average F1 score for
Prabowo topic. Supervised classification method required a
corpus (words and the weight of positive value and negative
value of the words) to classify a tweet. This research was
used the SentiwordNet corpus in the English language to
classify a tweet while the tweet data that was used in this
study itself written in Indonesia language. This research has
already explored experimentally Indonesian-language corpus
(Barasa) to classify a tweet and found several issues namely
lack of the phyton and Github document for Indonesia
language and imbalance of sentiment weight for the words.
Since those two issues, this research used SentiwordNet
dictionary and translate Indonesia tweet into English.
Unfortunately, not all tweets were written in formal
Indonesian language. Almost all tweets wrote in informal
978-1-7281-3333-1/19/$31.00 ©2019 IEEE
2019 International Conference on Information Management and Technology (ICIMTech)
66
19-20 Au
g
ust 2019, Jakarta & Bali, Indonesia

Indonesian language. Translating the informal Indonesia
language into English was another challenge, due to no
Indonesian informal language dictionary available. In
addition, the accuracy of the classifier depends on the data
(the time period of data collection and the number of data).
The data was collected on the campaign period. Collecting
data in a different time period would produce a difference
accuracy result.
V. C
ONCLUSION AND
F
UTURE
W
ORK
This study developed sentiment analysis using
Sentiwordnet and machine learning approach (Naïve Bayes
classification) for Indonesia general election opinion from
the twitter content. The results showed that the Naïve Bayes
classifier model achieved high performance to analyze the
sentiment content from Twitter on Indonesia language for
Indonesia general election opinion (Joko Widodo and
Prabowo topic). A detailed workflow has been introduced
that includes four main steps namely Data pre-processing,
Determine class label with SentiWordNet, Sentiment
analysis classification and Evaluate the result. Operationally
the sentiment analysis classification was performed by
cleansing and normalization process then translated tweet
into English through Google translate and Cloud Translation
API from Google Cloud Platform. Split text, Lemmatization
split text and SentiWordNet were employed to determine the
class label for a tweet. Tweet sentiment classification
performed using naive Bayes classifier. The result of this
research could be used as an alternative text analysis tools
instead of a survey approach for Indonesia general election
topic (presidential election) topic and could be considered as
an input for the campaign strategy.  ǡ  
         
     Ǥ    
 
ǦǤ   
       
      
ǦǤ
A
CKNOWLEDGMENT
We appreciatively admit the support from Research and
Technology Transfer office (TO), Bina Nusantara University
which helped the funding for this paper.
R
EFERENCES
[1] I. Adiyana and R. F. Hakim, “Implementasi text mining pada mesin
pencarian Twitter untuk menganalisis topik – topik terkait KPK dan
Jokowi”, Prosiding Seminar Nasional Matematika dan Pendidikan
Matematika UMS 2015, pp. 570-581, Yogyakarta: Universitas
Muhammadiyah Surakarta, 2015.
[2] W. K. Pertiwi, “Riset ungkap pola pemakaian medsos orang
Indonesia”, Kompas, March 1, 2018. [Online]. Available: Kompas,
https://tekno.kompas.com. [Accessed October 14, 2018].
[3] B. Marr, “How much data do we create every day? The mind-blowing
stats everyone should read”, May 21, 2018. [Online]. Available:
https://www.forbes.com. [Accessed October 13, 2018].
[4] Yudhianto, “Just a day, 2.5 quintillion bytes are created around the
world,” Dec. 10, 2016. [Online]. Available:
https://inet.detik.com/consumer/d-3367959/cuma-sehari-25-
quintillion-byte-tercipta-di-seluruh-dunia. [Accessed: Oct, 13, 2018].
[5] F. Thia, “An overview of future data trends that have an impact on
business,” Aug. 21, 2018. [Online]. Available:
https://sains.kompas.com/read/2018/08/21/205840823/gambaran-tren-
masa-depan-data-yang-berdampak-pada-bisnis. [Accessed: Oct, 13,
2018].
[6] M. V. Mäntyala, D. Graziotin, and M. Kuutilaa, “The evolution of
sentiment analysis—A review of research topics, venues, and top
cited papers”, Computer Science Review, Volume 27, pp. 16-32, Feb.
2018.
[7] X. Fang and J. Zhan, “Sentiment analysis using product review data”,
Journal og Big Data, 2:5, pp. 1-14, 2015.
[8] X. Zou, J. Yang, and J. Zhang, “Microblog sentiment analysis using
social and topic context”, Plos One, 13:2, pp. 1-24, 2018.
[9] B. Naiknaware, B. Kushwaha, and S. Ka wathekar, “Social Media
Sentiment Analysis using Machine Learning Classifiers”,
International Journal of Computer Science and Mobile Computing,
Vol.6 Issue.6, pp. 465-472, Jun. 2017.
[10] J. Hausler, J. Ruscheinsky, and M. Lang, “News-based sentiment
analysis in real estate: a machine learning approach”, Journal of
Property Research, Vol. 35, No. 4, pp. 344–371, 2018.
[11] E. F. Kusuma, “Bagaimana peran Twitter mempengaruhi politik
Indonesia?”, June 16, 2015. [Online]. Available: Inet Detik,
https://inet.detik.com. [Accessed: October 16, 2018].
[12] S. Adi, M. Wulandari, A. K. Mardiana, and A. Muzakki, “Survei:
topik dan tren analisis sentimen pada media online”, Seminar
Nasional Teknologi Informasi dan Multimedia 2018, pp. 55 – 60,
Yogyakarta: Universitas AMIKOM Yogyakarta., 2018.
[13] S. Saad and B. Saberi, “Sentiment Analysis or Opinion Mining: A
Review”, International Journal on Advanced Science, Engineering
and Information Technology, Vol. 7, No 5, pp. 1-7, 2017.
[14] J. D. Novakovi, A. Veljovi, S. S. Ili, Z. M. Papi. “Evaluation of
Classification Models in Machine Learning”, Theory and
Applications of Mathematics & Computer Science, Vol 7, No 1, pp.
39-46, 2017.
[15] W. Medhat, A. Hassan, and H. Korashy, “Sentiment analysis
algorithms and applications: A survey”, Ain Shams Engineering
Journal, Vol. 5, Issue 4, pp. 1093–1113, 2014.
[16] A. Yadollahi, A. G. Shahraki, and O. R. Zaiane, “Current state of text
sentiment analysis from opinion”, ACM Computing Surveys, Vol. 50,
No. 2, Article 25, pp. 1-33, 2017.
[17] D. Tsarev, M. Petrovskiy, and I. Mashechkin, “Supervised and
Unsupervised Text Classification via Generic Summarization”,
International Journal of Computer Information Systems and Industrial
Management Applications, Volume 5, pp. 509-515, 2013.
[18] S. Symeonidis, “5 Things You Need to Know about Sentiment
Analysis and Classification”, Democritus University of Thrace,
[Online]. Available: KDnuggets,
https://www.kdnuggets.com/2018/03/5-things-sentiment-analysis-
classification.html. [Accessed April 12, 2019]
[19] S. Qaiser and R. Ali, “Text Mining: Use of TF-IDF to Examine the
Relevance of Words to Documents”, International Journal of
Computer Applications, Volume 181, No.1, pp. 1-5, Jul. 2018.
[20] S. Baccianella, A. Esuli, and F. Sebastiani, “SentiWordNet 3.0: An
Enhanced Lexical Resource for Sentiment Analysis and Opinion
Mining”, Proceedings of the International Conference on Language
Resources and Evaluation, LREC 2010, 17-23 May 2010, Valletta,
Malta, pp. 2200-2204, 2010.
[21] G. Kaur and E. N. Oberai, “A Review Article On Naïve Bayes
Classifier With Various Smoothing Techniques”, International
Journal of Computer Science and Mobile Computing, Vol. 3, Issue.
10, pp. 864 – 868, Oct. 2014.
978-1-7281-3333-1/19/$31.00 ©2019 IEEE
2019 International Conference on Information Management and Technology (ICIMTech)
67
19-20 Au
g
ust 2019, Jakarta & Bali, Indonesia