IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 14, No. 2, April 2025, pp. 1616~1624

ISSN: 2252-8938, DOI: 10.11591/ijai.v14.i2.pp1616-1624  1616

A proposed approach for plagiarism detection in Myanmar

Unicode text

Sun Thurain Moe1, Khin Mar Soe1, Than Than Nwe2

1
Faculty of Computer Science, University of Computer Studies, Yangon, Myanmar
2
Faculty of Information Science, University of Information Technology, Yangon, Myanmar

Article Info ABSTRACT

Article history: Around the world, with technology that improves over time, almost
everyone can access the internet easily and quickly. With the increase in the
Received May 2, 2024 use of the internet, the plagiarism of information that is easily available on
Revised Oct 27, 2024 the internet has also increased. Such plagiarism seriously undermines
Accepted Nov 14, 2024 originality and ethical principles. In order to prevent these incidents, there is
plagiarism detection software for many countries and languages, but there is
no plagiarism detection software for the Myanmar language yet. In an
Keywords: attempt to fill that gap, this study proposed a deep learning model with
Rabin-Karp hash code and Word2vec model and built a plagiarism detection
Deep learning system. Our deep learning model was trained by randomly obtaining
Myanmar Unicode information from Myanmar Wikipedia. According to the experiments, our
Natural language processing proposed model can effectively detect plagiarism of educational content and
Plagiarism detection information from Myanmar Wikipedia. Moreover, it is possible to
Syllables segmentation distinguish plagiarized texts by rearranging words or substituting words with
some synonyms. This study contributes to a broader understanding of the
complexities of plagiarism in the Myanmar academic area and highlights the
importance of measures to effectively prevent plagiarism. It maintains the
credibility of education and promotes a culture that values originality and
intellectual integrity.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Sun Thurain Moe
Faculty of Computer Science, University of Computer Studies
D2, Room (608), Mindama Pyin Nyar Yeik Thar, Yangon, Myanmar
Email: [email protected]

1. INTRODUCTION
In the fields of literature and journalism, including various academic areas, the submission or
copying of intellectual property, which is someone else's efforts, without providing a reference or credit to
the original owner is gradually increasing, and it is becoming a challenge for various fields. The rapid and
easy access to vast amounts of information on the internet makes plagiarism attractive, and plagiarism
detection methods struggle to keep up with the growth of technologies such as artificial intelligence used in
plagiarism. The most advanced plagiarism detection systems available today use complex machine learning
and natural language processing techniques to find syntactic and semantic patterns in text. However, there is
a gap that needs to be filled, and that is the lack of proper application of these developments to Myanmar
Unicode text.
Plagiarism detection is an ever-evolving field within natural language processing, driven by the
increasing complexity of text and the sophisticated methods employed by those attempting to plagiarize.
Researchers have continuously explored various algorithms and techniques to improve the accuracy and
effectiveness of plagiarism detection systems. There are many different approaches in this sector, from rule-

Journal homepage: http://ijai.iaescore.com

Int J Artif Intell ISSN: 2252-8938  1617

based algorithms to advanced deep learning models, each contributing uniquely to improving detection
accuracy and efficiency.
Moe and Nwe [1] developed a highly accurate rule-based Myanmar syllable segmentation (MSS)
algorithm that achieves perfect segmentation accuracy on a large dataset of Myanmar Unicode text. This
algorithm's success underscores the potential of rule-based systems for handling specific linguistic
challenges. In the area of deep learning, El Mostafa and Benabbou [2] provided an extensive overview of
various propositions for plagiarism detection, highlighting the limitations of word granularity and Word2vec
methods in capturing the semantic nuances of sentences. Their study suggests the need for more sophisticated
models to accurately detect semantic plagiarism. Ali and Taqa [3] reviewed both traditional and modern
plagiarism detection techniques, concluding that intelligent and deep learning algorithms, which consider
lexical, syntactic, and semantic properties, outperform traditional methods, especially for large corpora. This
insight is pivotal for developing more effective plagiarism detection systems. Xiong et al. [4] introduced a
novel approach that integrates bidirectional encoder representations from transformers (BERT), an enhanced
artificial bee colony (ABC) optimization algorithm, and reinforcement learning (RL). This model addresses
imbalanced classification and has shown superior performance compared to existing models. Focused on
detecting plagiarism in social media content through a four-phase methodology involving data preprocessing, n-
gram evaluation, similarity analysis, and detection [5]. Their ensemble support vector machine based African
vulture optimization (ESVM-AVO) approach has demonstrated high accuracy and efficiency. Jambi et al. [6]
evaluated academic plagiarism detection methods using fuzzy multi-criteria decision-making (MCDM),
providing valuable recommendations for future systems. Eppa and Murali [7] proposed a multi-source
plagiarism detection method for C programming assignments, utilizing an attention-based model and density-
based spatial clustering of application (DBSCAN) clustering algorithm. Saeed and Taqa [8] developed an
application combining term frequency–inverse document frequency (TF-IDF) text encoding, natural
language processing, k-means clustering, and cosine similarity algorithms, while [9] enhanced plagiarism
detection using natural language processing and machine learning techniques, achieving impressive results
on benchmark datasets.
In cross-language plagiarism detection (CL-PD), Bouaine et al. [10] utilized Doc2vec embedding
techniques and a Siamese long short-term memory model, achieving outstanding accuracy and performance
metrics. Further advanced this field with transformer models and cross-lingual sentence alignment techniques
[11], [12]. AlZahrani and Al-Yahya [13] explored Arabic pretrained transformer-based models for authorship
attribution in Islamic law, fine-tuning models like ARBERT and AraELECTRA to achieve significant
results. Arabi and Akbari [14] proposed methods for detecting extrinsic plagiarism using pretrained networks
and WordNet ontologies, demonstrating high precision. Zouaoui and Rezeg [15] presented a multi-agent
indexing system for Arabic plagiarism detection, while El-Rashidy et al. [16] developed a system using
hyperplane equations for high accuracy, outperforming previous systems on standard datasets.
Elali and Rachid [17] examined artificial intelligence-based chatbots for detecting fabricated research, and
Anil et al. [18] compared the effectiveness of various plagiarism detection software on artificial intelligence
generated articles. Elkhatat et al. [19] evaluated artificial intelligence content detection tools' ability to
distinguish between human and artificial intelligence authored content, highlighting ongoing challenges in
this area. Foltýnek et al. [20] tested web-based text-matching systems, revealing that some systems can detect
certain plagiarized content but often misidentify non-plagiarized material. Muangprathub et al. [21] proposed
a formal concept analysis-based algorithm for document plagiarism detection, particularly effective with Thai
text collections. Tian et al. [22] introduced FPBirth for multi-threaded program plagiarism detection,
demonstrating significant performance improvements. Tlitova et al. [23] reviewed methods for identifying
cross-language borrowings in scientific articles, focusing on Russian-English pairs and the need for
specialized tools in this area. Ansorge et al. [24] presented a case study highlighting common errors in
paraphrased plagiarized texts, while Pal et al. [25] demonstrated improved accuracy in plagiarism detection
using natural language processing techniques, further advancing the field's capabilities.
These diverse studies collectively enhance our understanding and capability in plagiarism detection,
addressing various languages, contexts, and methodologies to ensure the integrity of academic and creative
works. The primary challenge addressed in this study is the lack of plagiarism detection tools for Myanmar
Unicode text. Without reliable plagiarism detection mechanisms, academic institutions and content creators
in Myanmar face difficulties in maintaining the integrity and originality of their work. Therefore, an
innovative method that makes use of the most recent developments in natural language processing and
machine learning is urgently needed in order to accurately detect plagiarism in Myanmar Unicode text.
Since the Myanmar language does not have a specific word cutoff, such as a space character, we
have worked step by step through complex preprocessing such as syllable segmentation, word tokenization,
stop word removal, and word embedding. Finally, we successfully built a very accurate and effective deep
learning model that can automatically identify text plagiarism cases across various topics on Myanmar
Wikipedia. To ensure the reliability and robustness of the model, we use a comprehensive evaluation process
A proposed approach for plagiarism detection in Myanmar Unicode text (Sun Thurain Moe)
1618  ISSN: 2252-8938

comparing its performance with established plagiarism detection techniques and manually annotated datasets.
In the sections on the following, we will explore the construction and preprocessing of the dataset, details of
the deep learning model architecture, and evaluation measures of this approach.

2. METHOD
In order to detect Myanmar Unicode plagiarism, input text, or documents are first processed to
separate syllables by the MSS algorithm. Then, using the pre-collected Myanmar word list and the longest
matching algorithm, it is converted into words. After that, stop words are removed from these words, and
sentences are segmented. Plagiarism detection will take a long time if the input sentences are compared with
all the texts on Myanmar Wikipedia. Therefore, keywords are extracted from the input sentences with the yet
another keyword extractor (YAKE) keyword extraction algorithm, and Wikipedia pages related to these
sentences are selected using the Wikipedia search application programming interface (API). Only the text
content is pulled from the selected Wikipedia pages, followed by syllable segmentation and word
tokenization. Then stop words are removed and segmented into sentences. Our proposed system design is
shown in Figure 1.

Input Text/Doc Text Content

Extraction

Text Processing Text Processing

Syllable Segmentation Syllable Segmentation

Wikipedia Search API

Word Tokenization Word Tokenization

Keyword Extraction

Stopword Removal Stopword Removal

Sentence Sentence
Segmentation Segmentation

Plagiarism Detection
Deep Fuzzy Jaro- Fuzzy Fuzzy Token
Fuzzy Cosine
Learning Winkler Levenshtein Sort

Paraphrase Counting and Similarity Calculation

Plagiarism Result

Figure 1. Myanmar Unicode plagiarism detection system

2.1. Myanmar syllable segmentation

The MSS algorithm extracts the Myanmar syllables from the input sentence based on the following
four rules [1]: i) If the ith syllabic element of input sentence is not a member of vowel_medial_group; ii) If

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1616-1624

Int J Artif Intell ISSN: 2252-8938  1619

the i-1th or ith or i+1th syllabic element of input sentence is not a consonant pairs symbol ‘္’; iii) If the i+1th
syllabic element of input sentence is not a consonant pairs symbol ‘္’; and iv) If the i+1th syllabic element of
input sentence is not a ‘Asat’ symbol ‘္’.

vowel_medial_group = [‘ေ္’, ‘္’, ‘္’, ‘္’, ‘္’, ‘္’, ‘္’, ‘္ ’, ‘္ ’, ‘္ ’, ‘္’, ‘္’, ‘္’, ‘္’, ‘ ္’, ‘္ ’]

If all the rules are correct, it has reached the start of the next Myanmar syllable. The example shown
below is a Myanmar sentence segmented into each Myanmar syllable using a syllable segmentation
algorithm.

Input Myanmar sentence:

ကယမင ကယခ င လမ တငအဆင မငလ ခ င အမင မည။

Output Myanmar syllables:

ကယ_မင _ကယ_ခ င _လမ _တင_အ_ဆင_ မင_လ _ခ င _အမ_င _မည_။

2.2. Word tokenization

We collected 46,837 Myanmar words and used a greedy longest-match-first technique to tokenize
Myanmar words. The algorithm picks the longest n-prefix of the remaining syllables that matches a word in
the pre-collected Myanmar word list. Syllables that are not included in the pre-collected Myanmar word list
are treated as unknown and dropped as stop words. Below is an example of converting Myanmar syllables
into Myanmar words.

Input Myanmar syllables:

ကယ_မင _ကယ_ခ င _လမ _တင_အ_ဆင_ မင_လ _ခ င _အမ_င _မည_။

Output Myanmar words:

ကယမင ကယခ င ၊ လမ ၊ အဆင မင၊ လ ခ င ၊ အမ၊ င မည

2.3. Stop words removal

There are no benchmark stop words that have been agreed upon in academia and specifically
defined in Myanmar natural language processing. In order to create a final stop word list, we combined our
ideas with the stop words previously discovered in Myanmar natural language processing research by other
researchers. Some of the stop words used in the proposed model are ‘၌’, ‘၍’, ‘၎င ’, ‘၏’, ‘က ေ ’,
‘ကျွ ေတ ’, ‘က မ’, ‘ ခင ’, ‘ ြင’, ‘ေ က င’, ‘ကတည က’ and all Myanmar digits [၀-၉].

2.4. Sentence segmentation

One of the fundamental processes in natural language processing is sentence segmentation. After
preprocessing the source and target corpora using the previously mentioned techniques, it separates a corpus
into discrete sentences. Sentence boundaries in the Myanmar corpus can be easily identified since the
Myanmar script uses a special character known as the "Pou ma" sign (။), similar to the full stop (.) in English,
to denote the end of a sentence.

2.5. Word2vec model

Google introduced Word2vec in 2013, which captures semantic and contextual similarities by
representing a continuous dense vector of words. A word embedding model can take massive textual corpora,
create a vocabulary dictionary, and generate a dense word embedding model that has a lower dimensionality
than bag-of-words models. There are two different model architectures, such as the continuous bag-of-words
(CBOW) and skip-gram models. In this work, we use the CBOW model and train with 45,399 sentences of
Myanmar Unicode text from 1,000 random Myanmar Wikipedia pages. Then the vector is used to represent
each word. In this phase, the entire text content of the Myanmar Wikipedia page is transformed into a matrix
of vectors, with each row representing a word. Figure 2 shows the visualization of our Word2vec model in

A proposed approach for plagiarism detection in Myanmar Unicode text (Sun Thurain Moe)
1620  ISSN: 2252-8938

the two-dimensional space of t-distributed stochastic neighbor embedding (t-SNE) with principal components
analysis (PCA), and each point represents a word.

Figure 2. Word2vec model

2.6. Rabin-Karp hash function

The Rabin-Karp algorithm was originally a checker that determines whether two strings (or
patterns) match. In this research, however, we employed the Rabin-Karp hashing approach to generate the
hash codes for Myanmar words and convert them to vectors. Unlike the Rabin-Karp technique, we did not
compare hashes according to linear time, but instead built a deep learning model and compared it. In this
work, we used the polynomial rolling hash and modular arithmetic methods defined in (1) to produce hash
codes for Myanmar words, ensuring that each Myanmar word received a unique hash code.

𝐻 = (𝑐1 ∗ 𝑏 𝑚−1 + 𝑐2 ∗ 𝑏 𝑚−2 + ⋯ + 𝑐𝑚 ∗ 𝑏 0 ) 𝑚𝑜𝑑 𝑄 (1)

Where H is the hash code, c is the integer American standard code for information interchange (ASCII) code
of the character in the word, b is the number of all Myanmar characters, m is the number of characters in the
word, and Q is a large prime number.

2.7. Deep learning model

After the preprocessing steps, we performed plagiarism detection on them using a deep learning
model. The deep learning model was trained using two different sentence vectorization techniques, such as
the Rabin-Karp rolling hash function and Word2vec. The training data includes 1,506 sentences randomly
taken from Myanmar Wikipedia pages. The training vectors are obtained after all the words from the training
sentences have been converted into hash code numbers, or Word2vec weights. The proposed model is
illustrated in Figure 3.
However, the length of the training vectors varies depending on how long the sentence is. We then
searched for the sentence with the most words and counted them, discovering that it contained almost
50 words. Each training vector was given a length of 50, and the blank spaces were filled with 1 s.
Concatenating the resulting 50-length vectors into 100-length vectors also includes adding the class label, as

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1616-1624

Int J Artif Intell ISSN: 2252-8938  1621

shown in Tables 1 and 2. When adding class labels, the class label is set to 1 if a vector repeats itself twice
and to 0 if it is adjacent to another randomly chosen vector. In this manner, we obtain a
3011×101-dimensional training dataset. It is important to clarify why each training vector's empty spaces
must be filled with 1s in this context.

Training Data Processing Phase

Text Document
Dataset Word Segmentation

Text Processing Dataset Creation

Vectorization
Syllable Segmentation (Word2Vec/Rabin-
Karp)

Sentence Training Dataset

Word Tokenization Segmentation Create Dataset

Stopword Removal Standardization

Deep Learning Training Phase

Deep Learning Model

Dense Layer(50, Input Dim=100,Activation=ReLu)

BatchNormalization Layer()
Dropout Layer(0.6)

Dense Layer(20, Activation=ReLu)

BatchNormalization Layer()
Dropout Layer(0.5)
Deep Learning Trained
Training Dataset
Model

Dense Layer(10, Activation=Softmax)

BatchNormalization Layer()
Dropout Layer(0.6)

Dense Layer(4, Activation=Sigmoid)

Dense Layer(2, Activation=Sigmoid)

Figure 3. Proposed deep learning model

Table 1. Training data (Rabin-Karp rolling hash)

Src W01 Src W02 … Src W49 Src W50 Tgt W01 Tgt W02 … Tgt W49 Tgt W50 Class
1207 2223 … 1 1 1207 2223 … 1 1 1
830 1146 … 1 1 830 1146 … 1 1 1
⁞ ⁞ … ⁞ ⁞ ⁞ ⁞ … ⁞ ⁞ ⁞
830 1146 … 1 1 236 467 … 1 1 0
455 920 … 1 1 515 1207 … 1 1 0

A proposed approach for plagiarism detection in Myanmar Unicode text (Sun Thurain Moe)
1622  ISSN: 2252-8938

Table 2. Training data (Word2vec)

Src W01 Src W02 … Src W49 Src W50 Tgt W01 Tgt W02 … Tgt W49 Tgt W50 Class
-0.01706486 -0.01323607 … 1 1 -0.01706486 -0.01323607 … 1 1 1
-0.01706486 -0.01334877 … 1 1 -0.01706486 -0.01334877 … 1 1 1
⁞ ⁞ … ⁞ ⁞ ⁞ ⁞ … ⁞ ⁞ ⁞
-0.01706486 -0.01334877 … 1 1 -0.03069331 -0.01706486 … 1 1 0
-0.01620913 -0.00342125 … 1 1 -0.01583702 -0.01572607 … 1 1 0

Checking for identity between the source string and the target string is the primary goal of
plagiarism detection. We aim to detect any instances of plagiarism in Myanmar Wikipedia articles. The entire
collection of articles on Myanmar Wikipedia will need to be used as training data if we choose to use a deep
learning model, which is what we typically do for this kind of requirement. There are many challenges
involved in doing this, including training data extraction, model training time, and hardware resources. Due
to the existence of these challenges, our deep learning model was separated from traditional operation models
and purposefully built as a probabilistic model based on weights similar to a logistics regression.
To maintain the weights of our trained model, we filled all empty spaces in the training vectors with
1 s. Our deep learning model does not use convolutional layers because our training dataset has a 1D
structure. It has only 5 dense, fully connected layers, and rectified linear unit (ReLU), softmax, and sigmoid
are used as activation functions. Using the holdout method, 20% of the 3,011 training datasets were divided,
and 602 were used as testing data for the model evaluation. Our proposed model has a 98% accuracy rate for
detecting plagiarism, as shown in Table 3.

Table 3. Results of proposed model

Precision Recall F1-score Support
Unmatched 1 0.96 0.98 302
Matched 0.96 1 0.98 300

Accuracy 0.98 602

Macro avg 0.98 0.98 0.98 602
Weighted avg 0.98 0.98 0.98 602

3. RESULTS AND DISCUSSION

As mentioned in subsection 2.7, we tested our proposed deep learning model using two different
vectorization techniques. The results indicated that the deep learning model trained with sentence
vectorization using the Rabin-Karp rolling hash function provided more accurate plagiarism detection results
than the model trained with Word2vec sentence vectorization. Our experiment is the first in the field of
Myanmar Unicode plagiarism detection, with no existing methods available for comparison. Plagiarism
detection techniques used in other languages, including English, are not applicable to Myanmar Unicode.
To evaluate the performance of our proposed model, we compared the results with those obtained
using well-known fuzzy string matching methods. The experiment involved 500 randomly selected sentences
from Myanmar Wikipedia, containing nearly 3,000 paraphrases. These sentences were tested for direct
copying and paraphrasing plagiarism, where word syntax was reversed. The results of this test are presented
in Table 4.
The experiment revealed that our proposed method, which combines a deep learning model with
Rabin-Karp sentence vectorization, produced the best results. However, our method still has some
weaknesses. As the first research for Myanmar Unicode plagiarism detection, focusing on direct copying and
pasting, the results shown in Table 4 are promising. Nevertheless, the method has limitations in detecting
other types of plagiarism, such as outlining and summarizing, where only the concept or idea is taken.
Further research is needed to develop adaptive methods for detecting these types of plagiarism.

Table 4. Experimental result

Method Similarity score (%)
Complete plagiarism Paraphrasing plagiarism
DL(Rabin-Karp) 95.6 95.6
DL(Word2Vec) 94.1 94.1
Fuzzy Jaro-Winkler 91.5 88.6
Fuzzy Levenshtein 92.4 59.4
Fuzzy Token Sort 91.9 91.5
Fuzzy Cosine 90.8 91.2

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1616-1624

Int J Artif Intell ISSN: 2252-8938  1623

4. CONCLUSION
Our study marks a significant step forward in the field of Myanmar Unicode plagiarism detection.
By testing a deep learning model trained with two different vectorization techniques, we found that sentence
vectorization with the Rabin-Karp rolling hash function provided superior accuracy compared to the
Word2vec-based approach. This experiment, the first of its kind for Myanmar Unicode, highlights the
limitations of applying plagiarism detection methods developed for other languages. While our method
showed promising results in detecting direct copying and paraphrasing, it still faces challenges in identifying
more complex forms of plagiarism, such as outlining and summarizing. Future research should focus on
developing adaptive methods to address these challenges, enhancing the robustness and accuracy of
plagiarism detection in Myanmar Unicode. In the future, such progress can be extended and significantly
contribute to maintaining academic and professional integrity while encouraging originality and creativity in
written works.

ACKNOWLEDGEMENTS
We would like to express our heartfelt gratitude to all those who contributed to the successful
completion of our publication paper. Additionally, we would like to express our gratitude for the support and
resources provided by the University of Computer Studies, Yangon, Myanmar. Their contributions were
essential to the completion of this work.

REFERENCES
[1] S. T. Moe and T. T. Nwe, “An algorithm for Myanmar syllable segmentation based on the official standard Myanmar Unicode
text,” in 2023 IEEE Conference on Computer Applications (ICCA), Feb. 2023, pp. 6–10, doi:
10.1109/ICCA51723.2023.10181391.
[2] H. El Mostafa and F. Benabbou, “A deep learning based technique for plagiarism detection: a comparative study,” IAES
International Journal of Artificial Intelligence (IJ-AI), vol. 9, no. 1, p. 81, Mar. 2020, doi: 10.11591/ijai.v9.i1.pp81-90.
[3] A. Ali and A. Taqa, “Analytical study of traditional and intelligent textual plagiarism detection approaches,” Journal of Education
and Science, vol. 31, no. 1, pp. 8–25, Mar. 2022, doi: 10.33899/edusj.2021.131895.1192.
[4] J. Xiong et al., “Efficient reinforcement learning-based method for plagiarism detection boosted by a population-based algorithm
for pretraining weights,” Expert Systems with Applications, vol. 238, p. 122088, Mar. 2024, doi: 10.1016/j.eswa.2023.122088.
[5] S. V. Vadivu, P. Nagaraj, and B. A. S. Murugan, “Ensemble machine learning technique-based plagiarism detection over opinions
in social media,” Automatika, vol. 65, no. 3, pp. 983–991, Jul. 2024, doi: 10.1080/00051144.2024.2326383.
[6] K. M. Jambi, I. H. Khan, and M. A. Siddiqui, “Evaluation of different plagiarism detection methods: A fuzzy MCDM
perspective,” Applied Sciences, vol. 12, no. 9, p. 4580, Apr. 2022, doi: 10.3390/app12094580.
[7] A. Eppa and A. H. Murali, “Machine learning techniques for multisource plagiarism detection,” in 2021 IEEE International
Conference on Computation System and Information Technology for Sustainable Solutions (CSITSS), Dec. 2021, pp. 1–5, doi:
10.1109/CSITSS54238.2021.9683752.
[8] A. A. M. Saeed and A. Y. Taqa, “A proposed approach for plagiarism detection in article documents,” SinkrOn, vol. 7, no. 2, pp.
568–578, Apr. 2022, doi: 10.33395/sinkron.v7i2.11381.
[9] F. K. AL-Jibory and others, “Hybrid system for plagiarism detection on a scientific paper,” Turkish Journal of Computer and
Mathematics Education (TURCOMAT), vol. 12, no. 13, pp. 5707–5719, 2021.
[10] C. Bouaine, F. Benabbou, and I. Sadgali, “Word embedding for high performance cross-language plagiarism detection
techniques,” International Journal of Interactive Mobile Technologies (iJIM), vol. 17, no. 10, pp. 69–91, May 2023, doi:
10.3991/ijim.v17i10.38891.
[11] R. S. R. Raj and G. R. Ramya, “Detection of plagiarism in contextual meaning using transformer model and community detection
algorithm,” Smart Trends in Computing and Communications, 2023, pp. 777–795, doi: 10.1007/978-981-99-0838-7_67.
[12] T. Ter-Hovhannisyan and K. Avetisyan, “Transformer-based multilingual language models in cross-lingual plagiarism detection,”
in 2022 Ivannikov Memorial Workshop (IVMEM), Sep. 2022, pp. 72–80, doi: 10.1109/IVMEM57067.2022.9983968.
[13] F. M. AlZahrani and M. Al-Yahya, “A transformer-based approach to authorship attribution in classical Arabic texts,” Applied
Sciences, vol. 13, no. 12, Jun. 2023, doi: 10.3390/app13127255.
[14] H. Arabi and M. Akbari, “Improving plagiarism detection in text document using hybrid weighted similarity,” Expert Systems
with Applications, vol. 207, Nov. 2022, doi: 10.1016/j.eswa.2022.118034.
[15] S. Zouaoui and K. Rezeg, “Multi-agents indexing system (MAIS) for plagiarism detection,” Journal of King Saud University -
Computer and Information Sciences, vol. 34, no. 5, pp. 2131–2140, May 2022, doi: 10.1016/j.jksuci.2020.06.009.
[16] M. A. El-Rashidy, R. G. Mohamed, N. A. El-Fishawy, and M. A. Shouman, “An effective text plagiarism detection system based
on feature selection and SVM techniques,” Multimedia Tools and Applications, vol. 83, no. 1, pp. 2609–2646, Jan. 2024, doi:
10.1007/s11042-023-15703-4.
[17] F. R. Elali and L. N. Rachid, “AI-generated research paper fabrication and plagiarism in the scientific community,” Patterns, vol.
4, no. 3, Mar. 2023, doi: 10.1016/j.patter.2023.100706.
[18] A. Anil et al., “Are paid tools worth the cost? a prospective cross-over study to find the right tool for plagiarism detection,”
Heliyon, vol. 9, no. 9, Sep. 2023, doi: 10.1016/j.heliyon.2023.e19194.
[19] A. M. Elkhatat, K. Elsaid, and S. Almeer, “Evaluating the efficacy of AI content detection tools in differentiating between human
and AI-generated text,” International Journal for Educational Integrity, vol. 19, no. 1, Sep. 2023, doi: 10.1007/s40979-023-
00140-5.
[20] T. Foltýnek et al., “Testing of support tools for plagiarism detection,” International Journal of Educational Technology in Higher
Education, vol. 17, no. 1, Dec. 2020, doi: 10.1186/s41239-020-00192-4.
[21] J. Muangprathub, S. Kajornkasirat, and A. Wanichsombat, “Document plagiarism detection using a new concept similarity in
formal concept analysis,” Journal of Applied Mathematics, vol. 2021, pp. 1–10, Mar. 2021, doi: 10.1155/2021/6662984.

A proposed approach for plagiarism detection in Myanmar Unicode text (Sun Thurain Moe)
1624  ISSN: 2252-8938

[22] Z. Tian, Q. Wang, C. Gao, L. Chen, and D. Wu, “Plagiarism detection of multi-threaded programs using frequent behavioral
pattern mining,” International Journal of Software Engineering and Knowledge Engineering, vol. 30, no. 11-12, pp. 1667–1688,
Nov. 2020, doi: 10.1142/S0218194020400252.
[23] A. Tlitova, A. Toschev, M. Talanov, and V. Kurnosov, “Meta-analysis of cross-language plagiarism and self-plagiarism detection
methods for Russian-English language pair,” Frontiers in Computer Science, vol. 2, Oct. 2020, doi: 10.3389/fcomp.2020.523053.
[24] L. Ansorge, K. Ansorgeová, and M. Sixsmith, “Plagiarism through paraphrasing tools—the story of one plagiarized text,”
Publications, vol. 9, no. 4, Oct. 2021, doi: 10.3390/publications9040048.
[25] S. K. Pal, O. J. Raffik, R. Roy, V. B. Lalman, S. Srivastava, and B. Sharma, “Automatic plagiarism detection using natural
language processing,” in 2023 10th International Conference on Computing for Sustainable Global Development (INDIACom),
2023, pp. 218–222.

BIOGRAPHIES OF AUTHORS

Sun Thurain Moe is currently pursuing a Ph.D. degree at the University of

Computer Studies, Yangon, Myanmar. He received a Master's degree (M.I.Sc.) in information
science from the University of Computer Studies, Yangon. He has contributed significantly to
the field of Myanmar syllable segmentation, Myanmar plagiarism and credit scoring,
publishing three papers in IEEE. His research interests include Myanmar natural language
processing, machine learning, and action recognition. He can be contacted at email:
[email protected].

Dr. Khin Mar Soe received a Ph.D. (information technology) degree from the
University of Computer Studies, Yangon, Myanmar. She is a professor at the University of
Computer Studies, Yangon. Her research interests are in the areas of natural language
processing, part-of-speech tagging, machine translation, and Myanmar name entity
recognition. She can be contacted at email: [email protected].

Dr. Than Than Nwe received a Ph.D. (information technology) degree from the
University of Computer Studies, Yangon, Myanmar. She is a professor at the University of
Information Technology, Yangon. Her research interests are in the areas of information
retrieval, data mining, big data, and pattern recognition. She can be contacted at email:
[email protected].

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1616-1624