Chatbot System for Mental Health in Bahasa Malaysia

Abstract

Chatbot has been the driving force of modern communication for business, customer service and even mental healthcare. At the same time, there are not many research and project regarding mental health chatbots in Bahasa Malaysia. This project focuses on developing a chatbot application for mental healthcare in Bahasa Malaysia. This chatbot system is integrated with artificial intelligence and natural language processing. This chatbot utilize the feedforward neural network model to train the datasets. Apart from the backend of the application, Kivy and KivyMD are used to create the app's graphical user interface.

Full text

Journal of Integrated and Advanced Engineering
(JIAE)
Vol. 2, No. 2, September 2022: 135-146
http://asasijournal.id/index.php/jiae
http://doi.org/10.51662/jiae.v2i2.83
M. I. Ismael et al., Chatbot System for Mental Health in Bahasa Malaysia
135
Chatbot System for Mental Health in Bahasa
Malaysia
Muhammad Imran Ismael1, Nik Nur Wahidah Hashim1, Nur Syahirah Mohd Shah2, and Nur Syuhada
Mohd Munir3
1Department of Mechatronics Engineering, Kulliyyah of Engineering, International Islamic University Malaysia, Malaysia
2Department of Nutrition Sciences, Kulliyyah of Allied Health Sciences, International Islamic University Malaysia, Malaysia
3Counselling and Career Services Center, International Islamic University Malaysia, Malaysia
Abstract
Chatbot has been the driving force of modern communication for business, customer
service and even mental healthcare. At the same time, there are not many research
and project regarding mental health chatbots in Bahasa Malaysia. This project
focuses on developing a chatbot application for mental healthcare in Bahasa
Malaysia. This chatbot system is integrated with artificial intelligence and natural
language processing. This chatbot utilize the feedforward neural network model to
train the datasets. Apart from the backend of the application, Kivy and KivyMD are
used to create the app's graphical user interface.
This is an open access article under the CC BY-SA license
Keywords:
Artificial Intelligence;
Chatbot;
Mental Health;
Natural Language Processing;
Article History:
Received: August 22, 2022
Revised: September 20, 2022
Accepted: September 24, 2022
Published: September 25, 2022
Corresponding Author: Nik
Nur Wahidah Hashim,
Mechatronics Department,
International Islamic University
Malaysia, Malaysia
Email:
niknurwahidah@iium.edu.my
INTRODUCTION
Through the mist of the COVID-19 outbreak, mass numbers of health issues, including
mental illness, rise rapidly, such as depression and anxiety disorders. In 2015, The National
Health and Morbidity Survey (NHMS), which the Ministry of Health organizes, declared that
one in three Malaysians have mental health issues, which amounts to 29.2% for those aged 16
and above.
Communication is essential for individuals to improve mental well-being as different
perspectives and answers can be obtained. With that in mind, some individuals may not enjoy
the act of conversing. Thus, this is where an application like a chatbot comes in handy. Most
chatbots use the concept of pattern matching using predefined words, which are stored in the
system's database. It is initially designed to act as a conversational agent for humans. However,
for a non-native English speaker, conversation with the chatbot can often be complex or
pointless.
This research conducted has eased the process for patients with mental health issues to
interact with a chatbot, particularly in Bahasa Malaysia, through the creation of okBot. okBot
is a chatbot system that is designed in Python for mental health conversations in Bahasa
Malaysia. This approach is the novelty presented in this study. With the help of machine
learning [1] and natural language processing (NLP) [2], this chatbot is capable of providing
accurate responses. Through this, individuals without proper medical attention can access some
guided therapy processes and acknowledge them.

Journal of Integrated and Advanced Engineering (JIAE)
Vol. 2, No. 2, September 2022: 135-146
136
M. I. Ismael et al., Chatbot System for Mental Health in Bahasa Malaysia
MATERIAL AND METHOD
Past Research
A systematic study of past research on mental health issues and the usage of chatbots in the
field is reviewed in this project. Depression or depressive disorder is a mood disorder which
provokes a constant feeling of sadness and loss of interest [3]. On the other hand, anxiety
disorders share the features of extreme and unreasonable fear and anxiety and related
behavioral disturbances. Therefore, it is necessary to have a background on mental illness
when creating the chatbot as it provides a sense of concept for the conversational flow of the
app.
This chatbot is inspired by many different mental health chatbots from past research. Among
them is Woebot, an automated conversational agent created to provide cognitive behaviour
therapy in the format of short and concise daily conversation and mood tracking [4]. Each
conversation with the app begins with a general inquiry of the user's mood with provided
responses.
In addition, there is another mental health chatbot that uses the native language, such as
Shim, which is in Swedish. Like Woebot, Shim is an automated conversational agent with
positive psychology and well-being themes such as expressing gratitude, practising kindness,
and replaying positive energy [5]. Chatbot as a system has the potential to be a useful tool for
a patient with mental disorders, specifically those who are hesitant to seek therapeutic help due
to stigmatization around it.
Data Collection and Intents Creation
For the purpose of creating a chatbot that utilizes a neural network in Python, the primary
approach is to gather the data. Without any data, there will not be any input towards the neural
network model. Neural network modelling will be explained in detail in later sections.
Mental Health Assessment (MHA)
This dataset consists of a list of individual expressions and their current state of emotion
throughout the weeks. In addition, this dataset contains the Patient Health Questionnaire (PHQ-
9) and General Anxiety Disorder (GAD-7) questionnaire scores from different individuals. The
PHQ-9 questionnaires are used as a measurement for severity of depression, while the latter is
for measurement of anxiety level.
Statistical Package for the Social Science (SPSS)
The SPSS datasets [6] have different structures for determining mental well-being. Among
the data useful for creating this chatbot is the mental well-being scores using the Short Warwick
Edinburg Mental well-being scale. Besides, similar to the MHA datasets, this dataset contains
a list of individuals expression throughout the day.
Counsel Chat
This dataset is obtained from the Counsel Chat website [7]. It contains a vastly different
topic that a professional therapist asked. The topics are depression, anxiety, relationships,
parenting, and trauma. To create this chatbot, questions asked on anxiety and depression are
used as sample training data, along with the respective responses from the therapist. The
selected data is first cleaned and translated into Bahasa Malaysia to prevent inconsistencies in
a dataset.

p-ISSN: 2774-602X e-ISSN: 2774-6038
M. I. Ismael et al., Chatbot System for Mental Health in Bahasa Malaysia
137
Intents JSON File
After gathering these datasets, an intent file is created in JSON format. This file will contain
all the tags, patterns, and responses of the conversational flow of the chatbot. The patterns and
responses are the user input and the responses given by the chatbot respectively while the tags
hold the classification of each. The bigger the data inside the intent file, the more accurate the
chatbot will be as there will be multiple different paths of conversations available for the chatbot
to work with when interacting with users.
Natural Language Processing
Natural Language Toolkit (NLTK) python package is used to implement NLP. NLP is one
of the ways for computer programs to understand human language [8, 9, 10]. Figure 1 shows
the pipeline for NLP, which consists of tokenization, lowercase conversion, stemming and
lemmatization, and removal of special characters.
The first technique, which is tokenization will tokenize sentences into individual words. For
instance, sentences like "I feel sad" will be tokenized into "I", "feel", "sad". Lowercase
conversion will simply change the cases in a text. Stemming and lemmatization turns
supported words into their root form, like organizations to organs. Finally, special characters
like "!" and "?" are removed.
Feedforward Neural Network
Neural networks are a set of algorithms modelled loosely after the human brain and are
designed to recognized patterns [11][12]. As a neural network requires numerical data as its
input, the processed text needs to be vectorized using the bag of words (bow) function, which
sets a word to either zero or one according to its occurrence in a particular pattern.
The vectorized words are then fed into the neural network, which is a feedforward neural
network. Feedforward neural networks are also artificial neural networks in which the
connections between units do not form a cycle [13, 14, 15]. This type of neural network is
chosen for its simplicity and because data in this chatbot is processed in one direction: from the
input to the output layer.
To achieve the highest accuracy possible in the neural networks, a lot of modification and
tweaking have to be done to change the hyperparameters of the neural network. The most
obvious one, which can be seen in Figure 2, is the number of hidden layers. The two hidden
layers in our neural network comprised 128 and 64 neurons, respectively. In addition to the
number of hidden layers, different activation functions inside the layers will have their usage.
This chatbot utilizes Rectified Linear Unit (ReLU) activation function for the hidden layers,
a piecewise linear function that will output the input directly whenever the result is positive.
Else, it will output zero. As for the output layer of the neural network, the Softmax activation
function will be used as it is suitable for single-label classification, which is important in
creating this chatbot. For instance, labelling a set of sentences into their appropriate tags.
Figure 1. NLP Preprocessing Pipeline

Journal of Integrated and Advanced Engineering (JIAE)
Vol. 2, No. 2, September 2022: 135-146
138
M. I. Ismael et al., Chatbot System for Mental Health in Bahasa Malaysia
Figure 2. Neural Network Model Visualization
Aside from that, dropout layers are implemented between the hidden layers and the output
layers. The dropout hyperparameters are the probability that some nodes are disconnected
during the training process. For example, Figure 2 shows that the input layer consists of 82
neurons while the output layer consists of 15 neurons generated from the number of classes and
patterns inside the JSON file.
In order to develop the best neural network mode, two deep learning frameworks are used
for this chatbot: TensorFlow and PyTorch. While the two frameworks implement the same
concepts of feedforward neural networks, they have different hyperparameters of the
optimizers, hidden layers, learning rate, epochs, and loss function. These hyperparameters are
changed gradually throughout training performance. If, for instance, the accuracy of the training
data is decreasing, increasing the number of hyperparameters may help improve the model's
accuracy. The condition is called underfitting.
Chatbot
Following the chatbot's training process, the program's main functionality, which is the
chatbot conversation, is created. A few notable functions used inside the Python file are the bag
of words function, which converts words into 0 or 1 vectors, as mentioned previously. Aside
from that, other function definitions are used to predict the classes from user input and predict
the correct responses with the implementation of an error threshold. This chatbot will first check
for classes with an error threshold lesser than 0.25 to give out the responses.
SQLite
SQLite is an embedded Structured Query Language (SQL) database engine with no separate
servers like most other SQL databases like MySQL [16]. The usage of SQLite database will
ease the usage of the chatbot on mobile devices like iOS and androids as it is a cross-platform
database. The database is implemented to store the conversation history between users and the
chatbot. This enables admins to analyze the chatbot performance along with the accuracy of
each intent's classification. okBot will store user's login information and chat history as in
Figure 3 and Figure 4.
The primary key in the chatbot database is the username generated through the registration
form, which is stored in the account table. On the other hand, the username in the chat table is
used as a foreign key that references the primary key in the account table. With this, the two
tables can be linked together to provide a one-to-many relation.

p-ISSN: 2774-602X e-ISSN: 2774-6038
M. I. Ismael et al., Chatbot System for Mental Health in Bahasa Malaysia
139
Figure 3. Accounts Table in okBot Database
Figure 4. Chat Table in okBot database
Graphical User Interface
Kivy and KivyMD are used to create the chatbot application's graphical user interface (GUI).
They are both python packages useful for developing portable user interfaces [16]. Kivy allows
the creation of a cross-platform application with multi-touch user interfaces. KivyMD, on the
other hand, is a collection of material design widgets for use with Kivy hence the name.
Although there are numerous ways to create a GUI, for example, using flutter or android studio,
Kivy is chosen as it is based on Python, which is applicable in this project. Besides, Kivy is
not an operating system (OS) dependent nor device dependent. Meaning that the code written
in Python can be used for different mobile devices such as iOS and Android.
The official documentation of Kivy and KivyMD provides a great way for inexperienced
developers to jump into the codebase [17][18]. Those with experience in HTML and CSS might
have a jumpstart in understanding the syntax for Kivy. Similar to web page designing using
HTML and CSS, there are multiple layouts to choose from when creating the chatbot interface.
The ones used in this application are float layout, box layout, and grid layout. The layout here
can be described as how pages and widgets are structured. Without them, the applications will
not be well-defined, and widgets might be all over the place. Kivy and KivyMD provide various
widgets for developers to choose from, which are the base building block of the application.
Widgets are able to receive events and respond accordingly.
Additional Features
The user experience with the application is enhanced through other features such as 1)
welcome page, 2) about section, 3) helpline section, 4) PHQ-9 test section, and 5) GAD-7 test
section. As the application boot up, the welcome page is shown with a login and registration
button so that the user’s credentials can be stored. The about section will go through briefly
what the app is and what is the main usage of the chatbot. The helpline page consists of a list
of emergency contacts. There will be instances where the chatbot will not be able to handle the
severity of an individual's issues due to the digital nature of the application. For the last two
sections, users can get an early assessment of their current mental health condition in terms of
depression and anxiety.

Journal of Integrated and Advanced Engineering (JIAE)
Vol. 2, No. 2, September 2022: 135-146
140
M. I. Ismael et al., Chatbot System for Mental Health in Bahasa Malaysia
Overview of Application
Figure 5 illustrates the overview of the whole application created in this project. The opening
page starts with the application page, which will prompt users for their login information or
registration queries. The app will then direct user to a chat screen where they can exchange
conversations with the chatbot. Then, through the neural network model that has been trained,
the chatbot will give out an accurate response based on the user's input sentence. Finally, the
navigation menu on the top left of the app will direct user to other sections of the app, as
mentioned in the previous section.
RESULTS AND DISCUSSION
This section will walk through the descriptive analysis of the MHA and SPSS dataset
gathered using Jupyter Notebook and the machine learning analysis of the chatbot. The GUI
flow of the chatbot will be illustrated near the end to provide a clear view of what has been
created for this particular mental health chatbot.
Descriptive Analysis of Mental Health Assessment
From the mental health assessment questionnaire, a distribution graph is created two identify
the average scores of depression and anxiety through PHQ-9 and GAD-7 tests. Figure 6
illustrates a pie chart of students' average PHQ-9 and GAD-7 scores at International Islamic
University Malaysia (IIUM). It can be observed that 61.6% of students participating in the
questionnaires have moderate to severe depression, while 38.4% range from minimal to mild
depression. The second pie chart shows that 46.1% of the respondents have moderate to severe
anxiety, while the other 53.9% range from minimal to mild anxiety.
Figure 5. okBot Flowchart

p-ISSN: 2774-602X e-ISSN: 2774-6038
M. I. Ismael et al., Chatbot System for Mental Health in Bahasa Malaysia
141
Figure 6. Average PHQ-9 and GAD-7 scores Among Students in IIUM
Descriptive Analysis of SPSS
A descriptive analysis is also made for the SPSS dataset provided. Figure 7 displays the
distribution data of students across IIUM. These three bar charts are illustrated to get a clearer
picture of their background when analyzing their mental well-being scores. The mental well-
being scores are calculated using the Short Warwick Edinburg Mental Well-Being Scale. As
for developing a mental health chatbot, it is important to know other underlying factors that
affect mental well-being.
Figure 8 and Figure 9 display a stacked bar chart which shows the relationship between self-
esteem and resilience against mental well-being. The obvious pattern that can be concluded is
that the higher the level of self-esteem and resilience, the higher the level of mental well-being.
In other words, students who can adapt better to their hardships have a higher chance of
developing better mental well-being. Figure 10 analyzes the correlation matrix between
numerical data in the SPSS dataset. The correlation matrix shows that total self-esteem and total
resilience have a positive correlation of 0.72 and 0.74 to mental well-being, respectively.
Figure 7. Relationship between Self-Esteem and Mental Well-Being

Journal of Integrated and Advanced Engineering (JIAE)
Vol. 2, No. 2, September 2022: 135-146
142
M. I. Ismael et al., Chatbot System for Mental Health in Bahasa Malaysia
Figure 8. Relationship between Self-Esteem and Mental Well-Being
Figure 9. Relationship between Resilience and Mental Well-Being
Figure 10. Correlation Matrix of SPSS Datasets

p-ISSN: 2774-602X e-ISSN: 2774-6038
M. I. Ismael et al., Chatbot System for Mental Health in Bahasa Malaysia
143
Natural Language Processing
Figure 11 illustrates the raw data inside the JSON file before the text-processing techniques
mentioned in section IV. In order to clean the initial patterns inside the JSON file, NLP
techniques such as tokenization, lemmatization and stemming, lower case conversion and
special case removal are used. Figure 12 displays the process text after NLP. At the current
stage of the chatbot, stemming and lemmatization cannot be achieved because of the lack of
support for the Malay language in the NLTK package. The alternative that can be taken is to
create a dictionary of stemmers.
Chatbot Evaluation
This chatbot is evaluated using two different deep learning frameworks: TensorFlow and
PyTorch. Although the two have a similar architecture of a neural network, the feedforward
neural network, different hyperparameters are used. For instance, neural network training using
TensorFlow is done through the Stochastic Gradient Descent (SGD) optimizer, while the latter
is done through the adaptive moment estimation (Adam) optimizer. Optimizer can be described
as the algorithm a neural network uses to update the training. Adam optimizer is one the most
popular and efficient optimizer for deep learning due to computational efficiency and its little
memory requirement [19][20].
Figure 13 and Figure 14 display the output of the training model, which display the accuracy
of each training module. TensorFlow has close to 100% accuracy with a 0.0026 loss percentage,
while PyTorch has a loss of 0.0004. The two trained models achieved similar results through
1000 epochs. The runtime performance will be measured during the conversational flow of the
user and chatbot.
Figure 11. Patterns Before Text-Processing
Figure 12. Patterns After Text-Processing

Journal of Integrated and Advanced Engineering (JIAE)
Vol. 2, No. 2, September 2022: 135-146
144
M. I. Ismael et al., Chatbot System for Mental Health in Bahasa Malaysia
Figure 13. TensorFlow Training Output
Figure 14. PyTorch Training Output
The two deep learning frameworks utilize different learning rates. Learning rate parameters
will determine how much the neural network will update each step through the neurons. For
example, in the TensorFlow training file, the learning rate is set to 0.01 while the latter is set
to 0.001. The implementation of both TensorFlow and PyTorch in this project is to understand
which frameworks are better suited for creating something like a chatbot. Table 1 displays the
performance of the chatbot through 50 user input tests.
PyTorch and TensorFlow are created by different companies: Facebook and Google.
Although Table 1. indicates that the accuracy of TensorFlow is better, PyTorch is often more
preferred to Python developers due to its Pythonic syntax, while having a lower learning curve
when compared to TensorFlow. Aside from that, different hyperparameters also affect the
accuracy of the chatbot. As far as we know, modifying the deep learning model is mostly a lot
of testing and trying an error from training performance.
Table 1: Performance Between TensorFlow and PyTorch
Model
Class
Average Word
Length
No Input
Patterns
Accuracy
(%)
TesnorFlow
15
5.829
82
90
PyTorch
15
5.829
82
80
CONCLUSION
Chatbots have become very important in multiple industries, such as online business,
customer support and mental health care. Using chatbots as therapy can help patients with
mental illnesses improve their conditions. As previously mentioned, chatbot usage for mental
health care in Bahasa Malaysia is not popular and less developed. This project introduced
okBot, a chatbot specifically in Bahasa Malaysia that supports individuals with mental health
problems. This research paper thoroughly analyses the approach taken to create an artificial
intelligent chatbot that utilises the concept of a neural network.
The approach to developing the chatbot began with data collection through three different
datasets. These help in creating the list of conversational flows in the chatbot and provide
training data for the neural network model. With the help of an open-source framework like
Kivy and KivyMD, a beautiful and modern design of GUI can be made. Apart from the usual
feature in a chatbot, okBot adds multiple features to improve the user experience with the
app. Among them is 1) the about page, 2) the helpline page, 3) the PHQ-9 test page, and 4) the
GAD-7 test page. The chatbot was evaluated with two different machine learning frameworks:
TensorFlow and PyTorch. For the database, okBot implement a structured query language using
SQLite.

p-ISSN: 2774-602X e-ISSN: 2774-6038
M. I. Ismael et al., Chatbot System for Mental Health in Bahasa Malaysia
145
As the technologies evolved, more and more improvements can be made to the chatbot
system. First and foremost, a feedback form can be created using the SMTP library, which
stands for Simple Mail Transfer Protocol, which can be used to send an email to any devices
that are connected with an SMTP listener. This allows chatbot users to provide feedback and
suggestions for future changes. Apart from that, the chatbot's overall architecture can be built
using a different type of neural network. A recurrent Neural Network (RNN) is another popular
network for developing a chatbot. While it has a higher complexity level when compared with
the simple feedforward neural network, it might be able to provide more personalized answers
and more specific responses as an RNN can memorise previous data points to help with
predictions.
REFERENCES
[1]
S. H. Jayady and H. Antong," Theme Identification using Machine Learning Techniques," Journal of
Integrated and Advanced Engineering (JIAE), vol. 1, no. 2, pp. 123-134, 2021, doi: 10.51662/jiae.v1i2.24
[2]
Z. P. Putera, M. D. Anasanti and B. Priambodo," Designing translation tool: between sign language to spoken
text on Kinect Time Series Data using Dynamic Time Warping," SINERGI, vol. 22, no. 2, pp. 91-100, 2018,
doi: 10.22441/sinergi.2018.2.004
[3]
R. L. Schalock, R. Luckasson & M. J. Tassé, "An Overview of Intellectual Disability: Definition, Diagnosis,
Classification, and Systems of Supports," American Journal on Intellectual and Developmental Disabilities,
vol. 126, no. 6, pp. 439-442, 2021, doi: 10.1352/1944-7558-126.6.439
[4]
K. K. Fitzpatrick, A. Darcy, and M. Vierhile, "Delivering Cognitive Behavior Therapy to Young Adults with
Symptoms of Depression and Anxiety Using a Fully Automated Conversational Agent (Woebot): A
Randomized Controlled Trial," JMIR Mental Health, vol. 4, no. 2, pp. e19, 2017, doi: 10.2196/mental.7785
[5]
K. H. Ly, A. M. Ly, and G. Andersson, "A fully automated conversational agent for promoting mental well-
being: A pilot RCT using mixed methods," Internet Interventions, vol. 10, no. October, pp. 39–46, 2017, doi:
10.1016/j.invent.2017.10.002
[6]
N. S. M. Shah, "Mental Well-being SPSS Datasets," 2022
[7]
N. Bertagnolli, "Counsel Chat: Bootstrapping High-Quality Therapy Data | by Nicolas Bertagnolli | Towards
Data Science," Towards Data Science, 2020. https://towardsdatascience.com/counsel-chat- bootstrapping-
high- quality-therapy-data-971b419f33da (accessed Jun. 17, 2022)
[8]
K. R. Chowdhary, Natural Language Processing. In: Fundamentals of Artificial Intelligence, Springer, New
Delhi: 2020, doi: 10.1007/978-81-322-3972-719
[9]
S. Meera & S. Geerthik, "Natural Language Processing," Artificial Intelligent Techniques for Wireless
Communication and Networking, 2022, doi: 10.1002/9781119821809.ch10
[10]
I. Lauriola, A. Lavelli and F. Aiolli, "An introduction to Deep Learning in Natural Language Processing:
Models, techniques, and tools," Neurocomputing, vol. 470, pp. 443-456, 2022, doi:
10.1016/j.neucom.2021.05.103
[11]
Q. Sun, C. Bai, H. Geng and B. Yu, "Deep Neural Network Hardware Deployment Optimization via
Advanced Active Learning," 2021 Design, Automation & Test in Europe Conference & Exhibition (DATE),
2021, pp. 1510-1515, doi: 10.23919/DATE51398.2021.9474100
[12]
A. de Santana Correia & E. L. Colombini, “Attention, please! A survey of neural attention models in deep
learning," Artificial Intelligence Review, 2022, doi: 10.1007/s10462-022-10148-x
[13]
S. D'Alfonso, "AI in mental health," Current Opinion in Psychology, vol. 36, pp. 112–117, 2020, doi:
10.1016/j.copsyc.2020.04.005
[14]
E. E. Lee et al., "Artificial Intelligence for Mental Health Care: Clinical Applications, Barriers, Facilitators,
and Artificial Wisdom," Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, vol. 6, no. 9, pp.
856-864, 2021, doi: 10.1016/j.bpsc.2021.02.001
[15]
N. Damij and S. Bhattacharya, "The Role of AI Chatbots in Mental Health Related Public Services in a
(Post)Pandemic World: A Review and Future Research Agenda," 2022 IEEE Technology and Engineering
Management Conference (TEMSCON EUROPE), 2022, pp. 152-159, doi:
10.1109/TEMSCONEUROPE54743.2022.9801962
[16]
C. Zhang and J. Yin, "Research on Security Mechanism and Forensics of SQLite Database," Advances in
Artificial Intelligence and Security, International Conference on Artificial Intelligence and Security (ICAIS
2021), 2021, pp. 614-629, doi: 0.1007/978-3-030-78618-2_51
[17]
G. Priyandoko, C. K. Wei and M. S. H. Achmad," Human following on ROS framework a mobile robot,"

Journal of Integrated and Advanced Engineering (JIAE)
Vol. 2, No. 2, September 2022: 135-146
146
M. I. Ismael et al., Chatbot System for Mental Health in Bahasa Malaysia
SINERGI, vol. 22, no. 2, pp. 77-82, 2018, doi: 10.22441/sinergi.2018.2.002
[18]
NN, "Welcome to KivyMD's documentation! — KivyMD 1.0.0.dev0 documentation."
https://kivymd.readthedocs.io/en/latest/ (accessed Jun. 17, 2022)
[19]
S. H. Cong and D. Kerzel," Allocation of resources in working memory: Theoretical and empirical
implications for visual search," Psychonomic Bulletin & Review, vol. 28, pp. 1093–1111, 2021, doi;
10.3758/s13423-021-01881-5
[20]
A. M. Jogdand and A. A. Magar, "Color therapy in mental health and well-being," International Journal of
All Research Education and Scientific Methods (IJARESM), vol. 10, no. 1, pp. 124– 130, 2022