Convolutional neural network modelling for autistic individualized education chatbot

IAES International Journal of Artificial Intelligence (IJ-AI)
Vol. 14, No. 1, February 2025, pp. 109~118
ISSN: 2252-8938, DOI: 10.11591/ijai.v14.i1.pp109-118  109
Journal homepage: http://ijai.iaescore.com
Convolutional neural network modelling for autistic
individualized education chatbot
Raseeda Hamzah1
, Nursuriati Jamil2
, Nor Diana Ahmad2
, Syed Mohd Zahid Syed Zainal Ariffin2
1
Computing Sciences Studies, College of Computing, Informatics and Mathematics, Universiti Teknologi MARA (UiTM), Melaka,
Malaysia
2
Computing Sciences Studies, College of Computing, Informatics and Mathematics, Universiti Teknologi MARA (UiTM), Shah Alam,
Malaysia
Article Info ABSTRACT
Article history:
Received Feb 26, 2024
Revised Aug 16, 2024
Accepted Aug 30, 2024
The traditional education system for autistic kids needs integration with
computer technology that embraces artificial intelligence to help school
instructors and management. An application that enables the teacher to
retrieve information from a trusted source is essential since the information
is only sometimes available on time. Thus, developing a chatbot application
that utilizes natural language processing can enhance the management of
autistic schools and will help individualized education for autistic students.
This research uses a deep learning model that utilizes a convolutional neural
network to develop a chatbot as a teaching assist tool for teachers. The
results show that the chatbot has achieved ˜0.03% loss when trained with
different epoch numbers. In terms of usability, the chatbot achieves mean
system usability scores of 80.48 ± 13.03. This may open opportunities for
more effective individualized education for students with special needs and
increase the potential to improve inclusive education for disabled students. It
is useful to include future actions that enable the simplification of the use of
this chatbot tool in a wide range of contexts. To close the education gap for
children with disabilities, chatbots could help people with communication
disabilities and could also significantly enhance the rate of communication.
Keywords:
Artificial intelligence
Autistic
Chatbot
Convolutional neural network
Education
This is an open-access article under the CC BY-SA license.
Corresponding Author:
Raseeda Hamzah
Computing Sciences Studies, College of Computing, Informatics and Mathematics
Universiti Teknologi MARA (UiTM)
110 off, St. Hang Tuah, Melaka Branch, Malaysia
Email: raseeda@uitm.edu.my
1. INTRODUCTION
Autism, often known as an autism spectrum disorder (ASD), is a severe condition characterized by
significant social interaction and communication difficulties. There is a broad spectrum of signs and abilities
that can range from mild inconvenience to severe disability requiring round-the-clock care in a specialized
facility [1]. Instructors usually have a hard time dealing with autistic students due to their lack of experience
in teaching and skills. Some schools may not have enough funding to hire professionals to overcome the
inexperience issues among instructors [2]. Most educators need more experience in teaching unique needs
students. Hence, the problem can be solved by having an application that enables the teacher to gain
information from a trusted source thus improving the individualized education system for autistic students.
The artificial intelligence (AI) field is a technology that constantly improves assisting daily human tasks. A
chatbot is a type of AI technology that can carry natural-sounding conversations for a variety of purposes,
thanks to its enormous vocabulary and wide range of conversational topics. Most internet banking

 ISSN: 2252-8938
Int J Artif Intell, Vol. 14, No. 1, February 2025: 109-118
110
applications [3], government officials [4], hospitals [5], universities [6], and other websites [7] are using
chatbots nowadays. However, autistic education sectors still implement conventional methods such as
manual books, materials, and physical interactions and communications. Currently, autistic education
teaching tools to assist teachers are still lacking. The current chatbots designed for autistic education mainly
focus on children’s usage. A chatbot for autistic kids’ usage was designed by [8]. Participants interacted with
the chatbots for 20 rounds of trials in their experiment to complete the evaluation. The results showed that the
chatbot can successfully attract students’ attention to understand it. The chatbot also showed how it could be
used in a conversation-based solution for Chinese kids with ASD. Another chatbot was developed by [9]
which is designed to diagnose achluophobia and autism using natural language processing (NLP) that
implements decision tree algorithms for disease severity detection. Not only can they save the time of an
expert diagnosis system, but they also ensure effectiveness. Similar research can be found in [10] where the
authors have developed a “Hear to help Chatbot” as an assistance tool for autistic kids to seek help. The
specific algorithm used in their AI chatbot has not been mentioned.
A chatbot based on forum data used by the autism online community was proposed in [11]. It
embedded a random forest algorithm for the classification task and concluded that the developed chatbot
could guess and reply correctly to the queries. It helps many users in finding information about autistics, at
least for the basic information. A shift from chatbot to robot was presented in [12]. The robot was introduced
to help the children complete their daily tasks and assist the teacher with educational activities. The kids and
teachers worked together to create the robot’s actions and how it fits into the school. The goal was to improve
the kids’ well-being. They used a mix of methods to look at how the robot was used throughout the study and
how its presence affected the kids’ well-being and the school’s environment. It can be observed that the robot
worked well in the school. It boosted the happiness of a select set of kids by encouraging and maintaining
regular contact with them. It also sparked a sophisticated discussion among students and faculty on the
benefits and drawbacks of this kind of social technology in the classroom.
Chatbots can be developed using several methods such as rule-based, retrieval-based, generative,
AI-assistant, social media, and hybrid. Rule-based chatbots follow a predefined set of rules and patterns to
respond. They typically use if-then statements or decision trees to determine the appropriate reply based on
specific keywords or patterns in the user’s input [13]. Retrieval-based chatbots store and retrieve predefined
responses from a database based on user inputs [14]. These responses are often prewritten and selected based
on the closest match to the user’s query. Generative chatbots employ more advanced techniques, such as NLP
and machine learning, to generate responses in real time. They learn from vast amounts of training data and
can generate contextually relevant and coherent responses [15]-[17]. AI-assistant chatbots, also known as
virtual assistants, integrate various AI technologies, including natural language understanding, speech
recognition, and machine learning. They aim to provide more comprehensive assistance by understanding
user intent, retrieving information from databases or application programming interface (APIs), and
performing tasks such as setting reminders, making reservations, or providing recommendations [18]. Social
media chatbots are designed specifically for interacting with users on social media platforms. They can
handle customer inquiries, provide information, deliver personalized content, and even facilitate transactions,
all within the social media messaging interface [19]. Hybrid chatbots combine different approaches, such as
rule-based systems, retrieval-based methods, and generative models, to leverage the strengths of each
technique. This allows for more flexible and robust conversations by switching between predefined responses
and dynamically generated ones [20]. The application of deep learning in chatbot development within the
corporate sector has shown a notable increase in recent times [21]. The effectiveness of chatbots is
continuously enhanced by their exposure to new discussions and user interactions, which the deep learning
models learn from [22], [23]. It is not debatable that AI and machine learning play a leading role in autistic
education. Apart from assisting the children and medical experts, there is plenty of improvement needed for
teachers and instructors to help grow the autistic children’s future. Therefore, this research will optimize AI
in the autistic education system in the hope that it will help all instructors and school management in their
daily tasks.
2. METHODOLOGY
Figure 1 shows this research’s overall flow, which contains raw data collection, data pre-processing,
chatbot development, and functionality testing. Overall, the research divides the phases into three phases:
system design, front-end development, and back-end development. The design process was facilitated by
developing design diagrams as shown in the following subsections. The system design includes the system
flowchart and use case diagram. The flowchart, as depicted in Figure 2, illustrates the overall flow of how the
chatbot operates and the system architecture in further depth.

Int J Artif Intell ISSN: 2252-8938 
Convolutional neural network modelling for autistic individualized education chatbot (Raseeda Hamzah)
111
Figure 1. Research flow
Figure 2. Chatbot flowchart
2.1. System design
One crucial aspect in the system design phase of chatbot development is to guarantee the efficient
transition between user queries and the chatbot's responses. In this research, the chatbot will ask the user if
they are satisfied with the response once they select the type of inquiry. If the user approves of the response,
the chatbot will offer three choices: rephrase the question, email headquarters, wait for a reply, or end the
conversation. Upon receiving an appropriate response, the user can decide whether to proceed with further
inquiries or conclude the session. Figure 3 displays the use-case diagram, a component of the system design
and development process, illustrating how a user interacts with the system. The use case diagram provides a
concise overview of the connections among the use cases, user, and system. Two cases originate from the
user, while one is generated by the system. The user will enter a query and then click the "Send" button. The
system will execute the algorithm to comprehend the user's intention and select the most dependable answer
from the trained dataset. Upon completion, the system will present the outcome in the chat. The user can
query the chatbot multiple times.
2.2. Front and back-end development
The backend development involves dataset design, convolutional neural network (CNN) model
development, and functionality testing. The development in the back end aims to facilitate the front-end
development with underlying data. Flask, a lightweight Python web framework with useful tools for building
Python online applications, will be used to develop the chatbot. Since it lets developers build a web program
in one Python file, it’s more flexible and easier for beginners. Flask may be expanded without a directory
structure or boilerplate code.
Dataset design is how the data is stored and used. Table 1 shows some examples of the raw dataset.
The data were collected manually by conducting interviews with the non-government organization (NGO)
representatives such as teachers and physiotherapists. The data were divided into two sections which are
educational and administrative.

 ISSN: 2252-8938
112
Figure 3. Chatbot use case diagram
Table 1. Data distribution
Types Examples Quantity
Educational 1. The child has poor gross motor skill
2. Is there a lesson plan for circle time?
3. How about toilet training?
120 sentences
840 words
40 repeated words
Administrative 1. How about teaching material?
2. How do I know about my staff profile?
3. Can you explain about staff leave?
150 sentences
955 words
138 repeated words
In addition to face-to-face meetings, internet meetings with the representatives were also used to
collect data. Before the interviewing sessions, ethical approval was sought and received. Observations on the
kids at the National Autistics Center were conducted to form a dataset that contains the frequently asked
questions (FAQ). The raw data that were used in this research went through a pre-processing stage such as
tokenization, stemming, and lemmatization. In the tokenization, the sentences were chunked into words.
Then, all the unnecessary marks such as punctuation marks, stop words, and fillers were filtered. Word
normalization was executed on the filtered tokenized word by using an NLP approach of stemming and
lemmatization. In stemming, the string was divided into substrings by utilizing specific rules. Stemming was
done to remove word affixes and suffixes thus reducing inflection in words to their root forms, hence
assisting in text, word, and document preparation for text normalization. On the other hand, lemmatization is
removing only inflectional endings and returning the lemma, which is the basic or dictionary form of a word
such as tense, case, voice, aspect, person, number, gender, and mood. Reducing inflectional words can avoid
redundancy in the NLP process. Although lemmatization is quite complex compared to stemming, it provides
the lexical and morphological of the words, which finally produces the dictionary form of the words.
The raw data, as shown in information gathering, were converted into a JavaScript object notation
(JSON) file, a form the algorithm can read. JSON is a lightweight format for storing and transporting data.
Figure 4 shows a snapshot of the JSON used for the chatbot. JSON is a standard data interchange format that
JSON file dictionary including tags, patterns, and responses [24], [25]. Traditionally, machine learning
modeling needs to go through a complex task such as data preparation and feature extraction before going
through different suitable modeling processes of training, testing, and validation. By using JSON, all the
hustles can be skipped. In the JSON dictionary, there will be one unique response for each tag, and there may
be multiple questions based on the tags. The tags were separated into two sections, each containing a single
answer with numerous intended user inputs. The tags also include keywords such as acquaintance and
greeting. The JSON file was trained using a CNN that is explained in the next sections.
Figure 4. JSON file for dataset design

113
Traditionally, machine learning modeling needs to go through a complex task such as data
preparation and feature extraction before going through different suitable modeling processes of training,
testing, and validation. By using JSON, all the hustles can be skipped. In the JSON dictionary, there will be
one unique response for each tag, and there may be multiple questions based on the tags. The tags are
separated into two sections, each containing a single answer with numerous intended user inputs. The tags
also include keywords such as acquaintance and greeting. The JSON file is trained using a CNN that is
explained in the next sections.
2.2.1. User interface design
The user interface (UI) design is one of the most important aspects of a system’s design to make it
appealing and user-friendly. Having a user interface design early in the project development process might
assist in avoiding mistakes and demonstrate how the system operates from the user’s perspective. The user
interface design is part of the front-end process that will have choices of buttons for the user to interact with
the chatbot. Figure 5 depicts a graphical user interface (GUI) that is offered for aesthetic purposes and to
make the dialogue more user-friendly. Based on Figure 5, when the user inputs a query and clicks the send
button the chatbot is connected to the trained dataset and able to post the response. The test was a success
since the chatbot sent a reply. Figure 5(a) lets you start the chatbot, Figure 5(b) lets you use a simple motion,
Figure 5(c) lets the user send a query, and Figure 5(d) lets the chatbot answer.
(a) (b) (c) (d)
Figure 5. Graphical user interface of AudiEBOT: (a) starting the chatbot, (b) a simple gesture, (c) sending a
query, and (d) Chatbot replies
2.2.2. Chatbot modelling using convolutional neural network algorithms
The CNN model was chosen to develop the chatbot because of its performance in predicting user
input compared to other conventional machine learning algorithms. The CNN is a widely recognized
architecture in the field of deep learning. It has been extensively employed in several domains, including
image processing and network intrusion detection [26]. One of the key advantages of CNNs is their ability to
facilitate deeper neural networks with significantly fewer parameters. Compared to other classification
methods, CNN requires substantially less pre-processing and can significantly learn all filters and features.
CNNs are multilayer perceptron (MLP) variants with convolutional layers. The convolutional layer
minimizes network complexity by applying a convolution function to the input and forwarding the result to
the next layer, analyzing a sentence/image at a time. CNN reduces complexity, allowing deeper networks to
handle more complicated input. In chatbot development, CNN takes a text as input and assigns important
parameters of learnable weights (w) and biases (b) to various features and objects in the text, allowing it to
differentiate between them. The CNN model that was used in this research consists of 3 hidden layers, each
with 10 neurons as shown in Figure 6. The batch size used in the CNN modeling is 10 with 0.001 learning
rate. The hidden neuron in each hidden layer acts the same as our brain which is influenced by the
organization of the visual cortex. The input size is related to the hidden size, and the hidden size is connected
to the number of classes, based on the relationship between the layers.
When the user provides input, the model initially tokenizes the information by dividing it into
tokens, which are smaller pieces of text. The tokens in this context consist of characters, words, and sub-
words. The tokens are subsequently serialized into a stream of 0 s and 1 s, a process known as serialization.
Then, it compares the input with the data from which the bot was trained and estimates the probability of that
input being associated with each tag. The pattern with the highest probability tag is considered and compared

 ISSN: 2252-8938
114
with a predetermined confidence threshold. If this tag’s probability exceeds the threshold, a random function
is used to display one of its responses on the user interface. The process is shown in Figure 7.
Figure 6. CNN with 3 hidden layers, I input layer, hn: hidden layers and o is the output layer
Figure 7. The system architecture
Multiple tests were set up for the modeling of CNN by testing a range of 500 to 4,500 epochs to
achieve minimal loss. The epochs were initially set at 500. However, the outcome appeared less than
promising. The number of epochs increased by 500 until it reached 4,500. It was discovered that epochs of
4,000 are optimal because they produced the lowest error rate. A predetermined confidence threshold
between 0.50 and 0.85 was established to compare the input with the data that was trained in the model. It
was discovered that when 0.7 was selected, too many inappropriate responses were displayed, and when 0.85
was selected, the CNN prediction was too strict. Consequently, the value of 0.75 was chosen. To measure the
accuracy of the chatbot, the accuracy of the chatbot was calculated using (1).
𝑎𝑐𝑐𝑢𝑟𝑎𝑐𝑦 =
𝐶𝑜𝑟𝑟𝑒𝑐𝑡 𝑐𝑙𝑎𝑠𝑠𝑖𝑓𝑖𝑐𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑖𝑛𝑠𝑡𝑎𝑛𝑐𝑒𝑠
𝑎𝑙𝑙 𝑖𝑛𝑠𝑡𝑎𝑛𝑐𝑒𝑠
(1)
3. RESULT AND DISCUSSIONS
The results present training and testing of the CNN model using specific pre-defined
hyperparameters and thresholds. The correctness of the chatbot is also evaluated numerous times to select the
most dependable number of cycles of the dataset that must be trained to achieve the lowest error. Several test
intents and their ground truth confidence percentages were prepared.
3.1. Modelling training result
Table 2 shows the training result of CNN modeling during the chatbot development that was done in
four folds. The results are also arranged according to the NLP threshold setting. Each epoch represents one
cycle of training, while loss is the proportion of train errors. The number of epochs is altered to determine
and minimize the number of errors. As described in the preceding section, multiple epochs were utilized to
train CNN. On average, four folds of training were done. The second fold denoted the lowest loss of 0.031.
3.2. Chatbot testing
To see the performance of the chatbot, two types of testing were done. The first was by using
threshold and the next one was using usability questionnaires gathered from [26]. A test case was constructed
to confirm that the system produced the intended result. The chatbot was tested for all potential flaws to
guarantee that the application could handle the problem.

115
Table 2. Training result of CNN modeling
Fold 1 Fold 2 Fold 3 Fold 4
Number of epoch Loss Number of epoch Loss Number of epoch Loss Number of epoch Loss
500 0.0069 500 0.0069 500 0.0069 500 0.0069
1,000 0.1744 1,000 0.1 1,000 0.2 1,000 0.1901
1,500 0.0002 1,500 0.0002 1,500 0.0002 1,500 0.0002
2,000 0 2,000 0 2,000 0 2,000 0
2,500 0.0876 2,500 0 2,500 0.0876 2,500 0.0876
3,000 0 3,000 0 3,000 0 3,000 0
3,500 0 3,500 0 3,500 0 3,500 0
4,000 0 4,000 0 4,000 0 4,000 0
4,500 0.1733 4500 0.1702 4,500 0.1715 4,500 0.1801
Average loss 0.0491 0.0308 0.0518 0.0516
3.2.1. Testing based on threshold
The chatbot considered a response valid if its confidence score is equal to or greater than 0.75. The
confidence score represents the model’s estimation of how confident it is in generating the response. Based
on the testing, when the 0.75 threshold was chosen, most test intent scored a confidence of more than 90%.
To find the optimum threshold value for the chatbot, we fed the bot with the test that has been annotated with
its correct responses or its correct ground truth value. Then the output was inspected based on the confidence
threshold at different levels of 0.50, 0.75, 0.80, and 0.85. There were about 150 intents tested for each of the
confidence thresholds. Results for the average accuracy against confidence thresholds are presented in
Table 3. It can be observed that the highest accuracy is at a confidence threshold of 0.80 which is about 85%.
Table 3. Test cases of confidence thresholds
Confidence threshold Average confidence score (%)
0.50 47
0.60 66
0.70 75
0.75 88
0.80 85
0.85 80
3.2.2. Chatbot usability questionnaires
A chatbot usability questionnaire is a set of questions designed to assess the user experience and
usability of a chatbot. It helps gather feedback from users to understand their satisfaction, ease of use, and
overall impression of interacting with the chatbot. The elements that are included in a chatbot usability
questionnaire as shown in Table 4. The questionnaires were ranked from strongly disagree to strongly agree,
with intermediate choices in between.
Table 4. Chatbot usability questionnaire
Questions
Strongly
disagree
Disagree Neutral Agree
Strongly
agree
Q1 The chatbot’s personality was realistic and engaging
Q2 The chatbot seemed too robotic
Q3 The chatbot was welcoming during the initial setup
Q4 The chatbot seemed very unfriendly
Q5 The chatbot explained its scope and purpose well
Q6 The chatbot gave no indication as to its purpose
Q7 The chatbot was easy to navigate
Q8 It would be easy to get confused when using the chatbot
Q9 The chatbot understood me well
Q10 The chatbot failed to recognize a lot of my input
Q11 Chatbot responses were useful, appropriate, and informative
Q12 Chatbot responses were not relevant
Q13 The chatbot coped well with any errors or mistakes
Q14 The chatbot seemed unable to handle any errors
Q15 The chatbot was very easy to use
Q16 The chatbot was very complex
The data collected from the surveys in Table 4 were analyzed using the system usability score (SUS)
following (2) as mentioned by [27].

 ISSN: 2252-8938
116
𝑆𝑈𝑆 =
1
𝑛
∑ 𝑛𝑜𝑟𝑚. ∑ {
𝑞𝑖, 𝑗 − 1, 𝑞𝑖, 𝑗𝑚𝑜𝑑2 > 0
5 − 𝑞𝑖, 𝑗, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
𝑚
𝑗=1
𝑛
𝑖=1 (2)
where m=10 is the number of questions, n=total number of subjects (questionnaires), 𝑞𝑖,𝑗=individual score
per question for each participant and norm=2.5. The results of the SUS can be categorized into percentile
ranges as shown in Table 5.
Table 5. Description of SUS grades
SUS grade Description of result
0-25 Worst Imaginable
>84.1 Best Imaginable
A group of 50 people who satisfied the requirements for being healthy adults were recruited as
volunteers to assess the usefulness of the chatbot. Figure 8 illustrates a boxplot that portrays the SUS
outcomes acquired from the assessment of the chatbot. The chatbot has achieved a mean SUS score of 80.48
± 13.03 with a median score of 85.95. The highest achieved score was 100.0, and the lowest recorded score
was 50. The chatbot's average score places it inside the top 93rd to 100th percentile range. Hence, it can be
inferred that the chatbot has a commendable degree of usability as determined by the conventional SUS
values.
Figure 8. Chatbot usability questionnaire scores
4. CONCLUSION
In this study, a chatbot is proposed as a solution to the problem of educational system management
for autistic children. The research focuses on how to assist teachers in managing their everyday
responsibilities while working with autistic students. It is an interactive system that provides pre-programmed
responses in response to questions posed by users. It frees customers from the constraints of time limits and
makes it possible for them to obtain responses quickly, making it an excellent substitute for conventional live
chat. The Python programming language will be utilized throughout the construction of the chatbot, which
will take place in the system gateway of the school. Users will have the ability to have a more natural
dialogue with the chatbot thanks to the implementation of deep learning, which has also been demonstrated
to have a greater level of accuracy when replying to user inquiries. In the not-too-distant future, the chatbot
will be incorporated into Telegram, an instant messaging service, and users can speak with one another
regarding autistic school management via Telegram. Users may also benefit from the chatbot system, notably
reducing the time necessary to conduct a physical consultation between teachers, physiotherapists, and the
upper management team.
ACKNOWLEDGEMENTS
The authors would like to thank the Ministry of Science, Technology and Innovation (Malaysia),
Yayasan Inovasi Malaysia and Universiti Teknologi MARA for their financial support to this project titled
smart personalized autism collaborative education system (SPACES): Collaborative Intelligent IEP Platform

117
(CIIP) under strategic research fund (SRF-APP). We would also like to thank the College of Computing,
Informatics and Mathematics, Universiti Teknologi MARA Shah Alam, Selangor for all the support. The
authors would also like to thank Muhammad Faris Mohamad Rosli from College of Computing, Informatics
and Mathematics, Universiti Teknologi MARA Shah Alam, Selangor for his efforts and involvement on
developing this project.
REFERENCES
[1] E. Long, S. Vijaykumar, S. Gyi, and F. Hamidi, “Rapid transitions: experiences with accessibility and special education during
the COVID-19 crisis,” Frontiers in Computer Science, vol. 2, 2021, doi: 10.3389/fcomp.2020.617006.
[2] N. Uithayakumar and N. M. Rabi, “Teachers’ readiness to accommodate autistic children in the classrooms at international
schools,” International Journal of Academic Research in Business and Social Sciences, vol. 10, no. 7, 2020, doi:
10.6007/ijarbss/v10-i7/7613.
[3] E. Mogaji, J. Balakrishnan, A. C. Nwoba, and N. P. Nguyen, “Emerging-market consumers’ interactions with banking chatbots,”
Telematics and Informatics, vol. 65, 2021, doi: 10.1016/j.tele.2021.101711.
[4] Y. Wang, N. Zhang, and X. Zhao, “Understanding the determinants in the different government ai adoption stages: evidence of local
government chatbots in China,” Social Science Computer Review, vol. 40, no. 2, pp. 534–554, 2022, doi:
10.1177/0894439320980132.
[5] J. Beaudry, A. Consigli, C. Clark, and K. J. Robinson, “Getting ready for adult healthcare: Designing a chatbot to coach
adolescents with special health needs through the transitions of care,” Journal of Pediatric Nursing, vol. 49, pp. 85–91, 2019, doi:
10.1016/j.pedn.2019.09.004.
[6] V. Oguntosin and A. Olomo, “Development of an E-commerce chatbot for a university shopping mall,” Applied Computational
Intelligence and Soft Computing, vol. 2021, 2021, doi: 10.1155/2021/6630326.
[7] C. C. Lin, A. Y. Q. Huang, and S. J. H. Yang, “A review of AI-driven conversational chatbots implementation methodologies and
challenges (1999–2022),” Sustainability, vol. 15, no. 5, 2023, doi: 10.3390/su15054012.
[8] X. Li, H. Zhong, B. Zhang, and J. Zhang, “A general Chinese chatbot based on deep learning and its’ application for children with
ASD,” International Journal of Machine Learning and Computing, vol. 10, no. 4, pp. 519–526, 2020, doi:
10.18178/ijmlc.2020.10.4.967.
[9] S. Mujeeb, M. Hafeez, and T. Arshad, “Aquabot: a diagnostic chatbot for achluophobia and autism,” International Journal of
Advanced Computer Science and Applications, vol. 8, no. 9, 2017, doi: 10.14569/ijacsa.2017.080930.
[10] “‘Hear’ to help Chatbot: Co-development of a chatbot to facilitate participation in tertiary education for students on the autism
spectrum and those with related conditions,” Autism CRC, Brisbane, USA, 2020.
[11] K. Mrini, M. Laperrouza, and P. Dillenbourg, “Building a question-answering chatbot using forum data in the semantic space,”
2018 Swiss Text Conference, 2018.
[12] S. Lemaignan, N. Newbutt, L. Rice, and J. Daly, “‘It’s important to think of pepper as a teaching aid or resource external to the
classroom’: A social robot in a school for autistic children,” International Journal of Social Robotics, vol. 16, no. 6, pp. 1083–
1104, 2024, doi: 10.1007/s12369-022-00928-4.
[13] S. A. Thorat and V. Jadhav, “A review on implementation issues of rule-based chatbot systems,” SSRN Electronic Journal, 2020,
doi: 10.2139/ssrn.3567047.
[14] D. Wang and H. Fang, “Length adaptive regularization for retrieval-based chatbot models,” ICTIR 2020 - Proceedings of the 2020
ACM SIGIR International Conference on Theory of Information Retrieval, pp. 113–120, 2020, doi: 10.1145/3409256.3409823.
[15] J. Kapočiute-Dzikiene, “A domain-specific generative chatbot trained from little data,” Applied Sciences, vol. 10, no. 7, 2020, doi:
10.3390/app10072221.
[16] M. A. Khadija, W. Nurharjadmo, and Widyawan, “Deep learning generative Indonesian response model chatbot for JKN-KIS,”
APICS 2022 - 2022 1st International Conference on Smart Technology, Applied Informatics, and Engineering, Proceedings, pp.
70–74, 2022, doi: 10.1109/APICS56469.2022.9918686.
[17] J. Kim, H. G. Lee, H. Kim, Y. Lee, and Y. G. Kim, “Two-step training and mixed encoding-decoding for implementing a
generative chatbot with a small dialogue corpus,” 2IS and NLG 2018 - Workshop on Intelligent Interactive Systems and Language
Generation, pp. 31–35, 2018, doi: 10.18653/v1/w18-6707.
[18] K. Deepika, V. Tilekya, J. Mamatha, and T. Subetha, “Jollity chatbot- a contextual AI assistant,” The 3rd International
Conference on Smart Systems and Inventive Technology, ICSSIT 2020, pp. 1196–1200, 2020, doi:
10.1109/ICSSIT48917.2020.9214076.
[19] D. Larbi, K. Denecke, and E. Gabarron, “Usability testing of a social media chatbot for increasing physical activity behavior,”
Journal of Personalized Medicine, vol. 12, no. 5, 2022, doi: 10.3390/jpm12050828.
[20] W. Maeng and J. Lee, “Designing a chatbot for survivors of sexual violence: exploratory study for hybrid approach combining
rule-based chatbot and ML-based chatbot,” 5th Asian CHI Symposium 2021, pp. 160–166, 2021, doi: 10.1145/3429360.3468203.
[21] M. Dhyani and R. Kumar, “An intelligent chatbot using deep learning with bidirectional RNN and attention model,” Materials
Today: Proceedings, vol. 34, pp. 817–824, 2019, doi: 10.1016/j.matpr.2020.05.450.
[22] E. Kasthuri and S. Balaji, “Natural language processing and deep learning chatbot using long short term memory algorithm,”
Materials Today: Proceedings, vol. 81, no. 2, pp. 690–693, 2021, doi: 10.1016/j.matpr.2021.04.154.
[23] M. Alruily, “ArRASA: channel optimization for deep learning-based Arabic NLU chatbot framework,” Electronics, vol. 11, no.
22, 2022, doi: 10.3390/electronics11223745.
[24] R. Parkar, Y. Payare, K. Mithari, J. Nambiar, and J. Gupta, “AI and web-based interactive college enquiry chatbot,” in
Proceedings of the 13th International Conference on Electronics, Computers and Artificial Intelligence, ECAI 2021, Jul. 2021,
doi: 10.1109/ECAI52376.2021.9515065.
[25] J. Batani, E. Mbunge, and L. Leokana, “A deep learning-based chatbot to enhance maternal health education,” in 2024
Conference on Information Communication Technology and Society, ICTAS 2024, 2024, pp. 7–11, doi:
10.1109/ICTAS59620.2024.10507149.
[26] M. Rohandi, N. Husain, and I. W. Bay, “Usability testing of intensive course mobile application using the usability scale system,”
ILKOM Jurnal Ilmiah, vol. 13, no. 3, pp. 252–258, 2021, doi: 10.33096/ilkom.v13i3.821.252-258.
[27] S. Holmes, A. Moorhead, R. Bond, H. Zheng, V. Coates, and M. McTear, “Usability testing of a healthcare chatbot: Can we use
conventional methods to assess conversational user interfaces?,” in Proceedings of the 31st European Conference on Cognitive
Ergonomics, Association for Computing Machinery, Sep. 2019, pp. 207–214, doi: 10.1145/3335082.3335094.

 ISSN: 2252-8938
118
BIOGRAPHIES OF AUTHORS
Raseeda Hamzah is a senior lecturer at the College of Computing, Informatics
and Media, Universiti Teknologi MARA (UiTM) Melaka Branch, Malaysia. Before joining
UiTM, she had 3 years of working experience in the telecommunication industry. She has a
Ph.D. in Information Technology and Quantitative Sciences from the Universiti Teknologi
MARA (UiTM). Her research interest is in pattern recognition, artificial intelligence, machine
learning, and the internet of things. She can be contacted at email: raseeda@uitm.edu.my.
Nursuriati Jamil is a professor from the School of Computing Sciences, College
of Computing, Informatics and Media, Universiti Teknologi MARA, Shah Alam. She
specializes in image and speech processing research and has been awarded international,
industry, and national grants for fundamental and social research. She is a senior member of
IEEE and has been involved in awarding student awards for IEEE Computer Society. She can
be contacted at email: liza_jamil@salam.uitm.edu.my.
Nor Diana Ahmad is a senior lecturer from the School of Computing Sciences,
College of Computing, Informatics and Media, Universiti Teknologi MARA, Shah Alam.
Before joining UiTM, she had 5 years of working experience in the information system
industry. She has a Ph.D. from the University of Leeds, United Kingdom. She specializes in
database technology research and her research interest is in database technology, information
retrieval, and NLP areas. She has been awarded national grants for science, technology, and
innovation research. She can be contacted at email: nordiana@tmsk.uitm.edu.my.
Syed Mohd Zahid Syed Zainal Ariffin obtained his B.Sc., M.Sc., and Ph.D. in
Computer Science from Universiti Teknologi MARA (UiTM). He is currently a senior lecturer
at the same university. His research interest areas are image processing, applied AI, and
instructional multimedia. He is a senior member of the IEEE Signal Processing Society. He
can be contacted at email: zahidzainal@uitm.edu.my.

Convolutional neural network modelling for autistic individualized education chatbot

More Related Content

Similar to Convolutional neural network modelling for autistic individualized education chatbot (20)

More from IAESIJAI (20)

Recently uploaded (20)

Convolutional neural network modelling for autistic individualized education chatbot