Automated detection of various types of plagiarism in academic papers of IT students

Abstract

The article explores the problem of automated detection of various types of plagiarism in academic works of IT students. Particular attention is paid to the more challenging forms of plagiarism to detect, especially those based on paraphrasing. An experiment was conducted using five popular plagiarism detection services. For analysis, a dataset was compiled consisting of 50 original texts and 712 paraphrased versions. A metric called Paraphrasing Detection Sensitivity (PDS) is proposed to quantitatively assess the ability of the services to identify paraphrasing plagiarism. The results revealed a significant variation in efectiveness among the systems, and conclusions were drawn regarding their practical applicability.

Keywords

• eol>plagiarism
• paraphrasing plagiarism
• automated processes
• paraphrasing detection sensitivity metric
• ethical academic practices
• academic originality assessment
• future IT Professionals
• vocational skill development
• automated processes
• digital rights law
• AI-powered information systems

1. Introduction

Information technologies play a key role in organizing labor, education, research, entertainment, and other aspects of everyday human activity [ 1, 2, 3 ]. They serve as a driving force that shapes the economic, technological, and scientific potential of a country.

In this regard, it is important to train future IT specialists, as they play an important role in the process of setting up and maintaining computer systems, participate in the development and further support of software, and influence the level of development and penetration of information technology in all spheres of human life [ 4, 5 ]. After graduation, future IT specialists should have a system of professional and key competencies, be ready to work in a team, have skills of self-learning, self-organisation and continuous professional growth.

The development of digital technologies, in addition to its obvious positive efects [ 6, 7 ], has negative aspects. With the development of digital technologies, students have gained wide access to information resources. This, on the one hand, increases the eficiency of the learning process, and on the other hand, creates preconditions for unauthorised copying of other people’s materials without proper acknowledgement. This practice undermines academic integrity, distorts assessment results and negatively afects the quality of training of future professionals. Therefore, detecting and preventing plagiarism is an important task for higher education institutions that requires efective solutions, in particular by automating the verification process.

The problem of automating the detection of plagiarism in student papers is not new. In particular, Durge et al. [ 8 ] describe the peculiarities of using the developed computer system for detecting plagiarism in computer programming. Gambo et al. [ 9 ] describe the GRAD-AI software tool, an automated tool for assessing computer programming assignments that combines automation with teacher involvement for accurate grading, timely feedback, and personalised support, enhancing the educational process. Sağlam et al. [ 10 ] proves the importance of human involvement in the process of automatic plagiarism detection. This paper presents a novel approach for automated plagiarism detection in modeling assignments that combines automated analysis with human inspection. The results show that we achieve a significantly higher detection rate for AI-generated attacks and a broader resilience than the state-of-the-art.

In recent years, humanity has been actively using artificial intelligence systems [ 11, 12, 13, 14, 15, 16 ], in particular generative artificial intelligence [ 17 ], to meet its daily needs. The educational process is no exception. The absence of ethical principles in generative AI and the ability of people to use it for their own benefit quite easily contribute to the use and abuse of AI-based tools by students of higher education institutions. Therefore, every year there is an increasing number of works devoted to the ethical aspects of using AI in the educational process. Among these works, it is worth highlighting papers [ 18, 19, 20 ].

Our study focuses on automating the process of detecting diferent types of plagiarism in the works of IT students. Therefore, it was important for our study to consider the types of academic plagiarism. Based on the analysis of works [ 21, 22 ], the following types of plagiarism were identified: complete plagiarism, direct plagiarism, accidental plagiarism, plagiarising yourself, paraphrasing plagiarism, source-based plagiarism, mosaic plagiarism, AI plagiarism.

This is not the first time that the authors of this article have studied the problem of detecting plagiarism in student papers. Thus, in [ 23 ], the authors investigated the peculiarities of using artificial intelligence systems to assess the complexity of an algorithm based on student code fragments. In [ 22 ], the authors developed a software application to automate the routine work of university professors in checking student papers for plagiarism and detecting it with the help of AI. The developed software application uses various APIs to search for plagiarism and fragments of text/code generated by AI.

The purpose of this study is to determine the efectiveness of existing automatic plagiarism detection systems in recognising diferent types of plagiarism in the works of IT students.

2. Methodology

Let us take a closer look at the types of plagiarism and the possibilities of their automatic detection. It is evident that complete plagiarism and direct plagiarism are the easiest types to detect using automated tools. In the case of complete plagiarism, a student or author entirely copies someone else’s work without making any changes, whereas direct plagiarism involves verbatim reproduction of individual text fragments without proper citation.

Accidental plagiarism occurs when a student unknowingly presents someone else’s ideas, text or code as their own due to a lack of knowledge about proper citation, poor paraphrasing skills or a misunderstanding of what exactly constitutes plagiarism. From the point of view of automatic detection, unintentional plagiarism is a separate issue. Typical checking systems may flag such cases in the same way as intentional plagiarism, without taking into account the author’s intent, as such systems evaluate the result, not the process.

From a technical point of view, detecting self-plagiarism is no diferent from detecting direct plagiarism, as both cases involve finding textual or structural similarities between documents. However, the key diference lies in the source of comparison: to detect self-plagiarism, it is needed to have access to the student’s historical work, including previously submitted term papers, labs or projects. Thus, efective detection of self-plagiarism is only possible if the algorithm can compare the new work not only with open sources, but also with the local internal archive of the educational institution.

Source-based plagiarism is a type of academic dishonesty in which a student misuses or intentionally distorts sources. Such plagiarism does not necessarily involve copying text, but rather manipulating references.

AI plagiarism is a type of plagiarism when a student submits as his or her own work text or code generated in whole or in part by artificial intelligence systems. Although such materials may be formally “unique” from the point of view of classical plagiarism detection systems, they are not the product of independent intellectual activity, which contradicts the principles of academic integrity. The main dificulty lies in the fact that AI-generated text may not have any matches with public sources, so automatic verification tools may not detect violations. In [ 22 ], the authors conclude that none of the services demonstrate good results in detecting AI plagiarism.

A special place among the types of plagiarism is occupied by paraphrasing plagiarism and mosaic plagiarism. Paraphrasing occurs when a student rewrites someone else’s text in his or her own words, preserving the main idea, structure, and content without proper reference to the source. Mosaic plagiarism is the creation of a text by collecting fragments from diferent sources with partial paraphrasing, changing grammatical structures, or replacing words with synonyms without proper citation. From the point of view of automatic detection, these types of plagiarism are much more dificult to identify than verbatim copying. This requires algorithms that take into account not only superficial but also semantic similarities between texts.

Studies show that paraphrasing is a common practice among students and, in some cases, even their primary strategy. At the same time, students tend to avoid deep comprehension of the original text, favoring superficial or mechanical paraphrasing, also known as patchwriting [ 24 ].

That is why in our work we paid attention to determining the quality of automatic paraphrasing detection. As part of the study, we conducted an experiment using a number of automated plagiarism detection services. We are not interested in the generalised quality of the services, but rather in the diference between their ability to detect complete plagiarism and direct plagiarism and their ability to detect paraphrasing plagiarism.

Two sets of textual data were prepared as part of the study. The first set (named TO (1)) contained fragments of student papers (term papers, bachelor’s and master’s theses) published on the Internet. Since these texts are publicly available, plagiarism detectors should recognise them as cases of direct plagiarism with a complete or almost complete match.

TO = {to| = 1 . . . } (1) where TO is a set of original (published) student texts, to is an -th text from TO set, = 1, . . . , , and is the total number of texts in this set.

For further comparative analysis, a set of sets (named TP (2)) of paraphrased versions was created for each text from the first set. Some of them were generated using artificial intelligence tools, while others were paraphrased manually.

TP = {{tp | = 1 . . . }| = 1 . . . }

TP = {TP| = 1 . . . } where TP is a set of sets of paraphrased versions of student texts, is a number of paraphrased versions for text to, = 1, . . . , , tp is a -th paraphrased version of text to, = 1, . . . , , = , . . . , .

Thus, each original text to ∈ TO corresponds to a subset TP:

TP = {tp | = 1 . . . }

The result of the to text plagiarism detection will be a certain value R(to)∈ [0 . . . 1]. The same applies to the values R(tp )∈ [0 . . . 1], which are the results of the plagiarism detection for texts tp . A value of 0 means that no plagiarism has been detected, and a value of 1 means that the entire text is plagiarised. An ideal plagiarism detector would return a value of 1 for all texts that participate in the experiment. (2) (3)

To evaluate the quality of the paraphrasing plagiarism detection, we ofer a metric of sensitivity to paraphrasing – paraphrasing determination sensitivity (PDS):

=1 =1 = 1 ∑︁ 1 ∑︁ |(to) − (tp )| (4)

From expression (4), it follows that PDS is also a normalised value, i.e. it falls within the interval [0...1]. A low PDS value means that the service responds almost equally to original and paraphrased texts (a value of 0 means that the response is the same). A high PDS value means that the service is much better at detecting complete plagiarism and direct plagiarism than paraphrasing plagiarism.

3. Results and discussion

We reviewed 37 plagiarism detection services. Among the available services, we chose those that demonstrate good results in detecting complete plagiarism and direct plagiarism, as there is no point in checking texts for paraphrasing plagiarism in cases where even basic forms of plagiarism are not detected efectively. The list of such services is as follows:

1. Paraphraser plagiarism checker [25].

2. Desklib plagiarism checker [ 26 ]. 3. Plagiarism Checker by JustDone [ 27 ]. 4. 1text.com Plagiarism Checker [ 28 ]. 5. Smart Plagiarism Checker by ExpertChat AI [ 29 ].

As part of the experiment, we selected 50 original student works (bachelor’s and master’s theses, course projects) that are publicly available. These student papers form the TO set ( = 50).

To generate the paraphrased texts, the authors used both artificial intelligence tools (with prior verification of the generated output) and manually created texts. For each original text, between 9 and 20 paraphrased versions were produced (9 ≤ ≤ 20). The total number of paraphrased texts is 712.

As an example, table 1 presents the values obtained by running all the services on one of the original texts, along with the results for its paraphrased versions. The values are expressed as percentages.

If only one of the above original texts had been used in the experiment, the values of the paraphrasing sensitivity metrics for each of the services would have been the same as those shown in table 2.

As a result of performing the verification procedure for each of the test texts from the TO and TP sets using each of the plagiarism detection services, a data set was obtained, on the basis of which the sensitivity of each of the studied services to paraphrased texts was calculated. This allows for a comparative analysis of their efectiveness. These results are presented in table 3.

The results show a significant diference between diferent services in terms of sensitivity to paraphrasing. The PDS values for ExpertChat and JustDone are close to 0, which means that they provide high-quality verification of paraphrased texts. The services from 1text.com and Desklib demonstrate good results only for direct plagiarism, but they are unable to recognise plagiarism in paraphrased texts. The Paraphraser service identifies a significant percentage of paraphrased text and can therefore be used as a support mechanism.

The total number of texts for which the plagiarism checking procedure was executed is calculated in expression (5): (5)

This exceeds the minimum sample size required ( ≈ 384) to estimate proportions with a 95% confidence level and a margin of error of ± 5%. Accordingly, the obtained sensitivity values can be considered statistically reliable with an error of no more than ± 3.7%. In addition, since all the texts in the study belong to the IT industry, this ensures high internal homogeneity of the data.

4. Conclusion

This study conducted a comprehensive analysis of the efectiveness of modern services for the automatic detection of plagiarism in student texts in IT. Particular attention is paid to the forms of plagiarism that are dificult to detect, such as paraphrasing.

An experimental dataset was created, consisting of 50 original texts and 712 paraphrased versions. Each text was checked using five popular plagiarism detection services.

In order to quantify the ability of services to recognise paraphrasing plagiarism, the PDS (paraphrasing detection sensitivity) metric was proposed. The results of calculating this metric showed a significant diference in sensitivity between diferent systems and confirmed the limited ability of most tools to detect plagiarism in transformed texts.

In the future, it is planned to investigate how the type of paraphrasing (lexical, syntactic, semantic or pattern) afects the efectiveness of plagiarism detection. Such an analysis will help identify weaknesses in the work of existing services and formulate recommendations for their improvement.

Declaration on Generative AI

The authors have not employed any generative AI tools.

References

[1] O. Malykhin , N. Aristova , N. Dichek , N. Dyka , Formation of top job skills of tomorrow among computer engineering and information technologies undergraduate students in the process of learning English, Vide . Tehnologija. Resursi - Environment, Technology, Resources 2 ( 2021 ) 249 - 254 . doi: 10 .17770/etr2021vol2. 6642 .

[2] O. Blavt , G. Iedynak, S. Pereverzieva , V. Holub , S. Melnyk , Increasing the reliability of test control using information technologies in inclusive physical education , Physical Education Theory and Methodology 23 ( 2023 ) 607 - 613 . doi: 10 .17309/tmfv. 2023 . 4 .16.

[3] N. V. Shumeiko , K. P. Osadcha , Application of artificial intelligence in higher education institutions for developing soft skills of future specialists in the sphere of information technology , Journal of Physics: Conference Series 2871 ( 2024 ) 012027 . doi: 10 .1088/ 1742 -6596/2871/1/012027.

[4] S. O. Leshchuk , I. A. Tverdokhlib , T. V. Pidhorna , D. I. Huska , A. A. Patyashina , Teaching programming for future information technology specialists based on end-to-end design of computer games , CEUR Workshop Proceedings 3949 ( 2025 ) 206 - 225 . URL: https://ceur-ws. org/ Vol- 3949 /paper17.pdf.

[5] S. V. Symonenko , N. V. Zaitseva , V. V. Osadchyi , K. P. Osadcha , V. S. Kruglyk , S. O. Sysoieva , Application of chatbots for enhancing communication skills of IT specialists , Journal of Physics: Conference Series 2871 ( 2024 ) 012026 . doi: 10 .1088/ 1742 -6596/2871/1/012026.

[6] T. A. Vakaliuk , D. S. Antoniuk , The use of digital technology in general secondary education in Ukraine: current state and future prospects , in: S. Papadakis (Ed.), Proceedings of the VIII International Workshop on Professional Retraining and Life-Long Learning using ICT: Personoriented Approach (3L-Person 2023 ), Virtual Event, Kryvyi Rih, Ukraine, October 25 , 2023 , volume 3535 of CEUR Workshop Proceedings, CEUR-WS.org , 2023 , pp. 17 - 31 . URL: https://ceur-ws. org/ Vol- 3535 /paper01.pdf.

[7] T. H. Kramarenko , V. M. Kramarenko , The use of digital technologies in education in the context of sustainable development of society , IOP Conference Series: Earth and Environmental Science 1415 ( 2024 ) 012013 . doi: 10 .1088/ 1755 -1315/1415/1/012013.

[8] T. Durge , A. Bansod , S. Deshmukh , S. Boke , C. Shewale , P. Wawage, AI-driven Student Task Management Platform With Plagiarism Checker For Submissions , in: Proceedings of the 5th International Conference on Emerging Systems and Intelligent Computing (ESIC 2025 ), 2025 , pp. 943 - 948 . doi: 10 .1109/ESIC64052. 2025 . 10962738 .

[9] I. Gambo , F.-J. Abegunde , O. Gambo , R. O. Ogundokun , A. N. Babatunde , C.-C. Lee , GRAD-AI : An automated grading tool for code assessment and feedback in programming course , Education and Information Technologies 30 ( 2025 ) 9859 - 9899 . doi: 10 .1007/s10639-024-13218-5.

[10] T. Sağlam , S. Hahner , L. Schmid , E. Burger, Automated Detection of AI-Obfuscated Plagiarism in Modeling Assignments, in: Proceedings of the 46th International Conference on Software Engineering: Software Engineering Education and Training , ICSE-SEET ' 24 , Association for Computing Machinery, New York, NY, USA, 2024 , p. 297 - 308 . doi: 10 .1145/3639474.3640084.

[11] O. M. Haranin , N. V. Moiseienko , Adaptive artificial intelligence in RPG-game on the Unity game engine , in: A. E. Kiv, S. O. Semerikov , V. N. Soloviev , A. M. Striuk (Eds.), Proceedings of the 1st Student Workshop on Computer Science & Software Engineering , Kryvyi Rih, Ukraine, November 30 , 2018 , volume 2292 of CEUR Workshop Proceedings, CEUR-WS.org , 2018 , pp. 143 - 150 . URL: https://ceur-ws. org/ Vol- 2292 /paper16.pdf.

[12] O. V. Korotun , T. A. Vakaliuk , A. M. Makhno , Tools for Teaching the R Programming Language to Bachelors of Computer Science in the Period of Distance Learning , in: E. Smyrnova-Trybulska , N.-S. Chen , P. Kommers , N. Morze (Eds.), E-Learning and Enhancing Soft Skills: Contemporary Models of Education in the Era of Artificial Intelligence , Springer Nature Switzerland, Cham, 2025 , pp. 309 - 330 . doi: 10 .1007/978-3- 031 -82243-8_ 18 .

[13] A. Bielinskyi , V. Soloviev , V. Solovieva , H. Velykoivanenko , Fuzzy time series forecasting using semantic artificial intelligence tools, Neuro-Fuzzy Modeling Techniques in Economics 2022 ( 2022 ) 157 - 198 . doi: 10 .33111/nfmte. 2022 . 157 .

[14] K. P. Osadcha , N. V. Shumeiko , Artificial intelligence, the labor market, and education for sustainable development: the points of intersection , IOP Conference Series: Earth and Environmental Science 1415 ( 2024 ) 012015 . doi: 10 .1088/ 1755 -1315/1415/1/012015.

[15] O. Y. Tarasova , V. S. Doroshko , Methodological foundations of teaching the basics of artificial intelligence to lyceum students , in: S. O. Semerikov , A. M. Striuk , M. V. Marienko , O. P. Pinchuk (Eds.), Proceedings of the 7th International Workshop on Augmented Reality in Education (AREdu 2024 ), Kryvyi Rih, Ukraine, May 14 , 2024 , volume 3918 of CEUR Workshop Proceedings , CEURWS.org, 2024 , pp. 240 - 249 . URL: https://ceur-ws. org/ Vol- 3918 /paper184.pdf.

[16] K. Bondar , O. S. Bilozir , O. Shestopalova , V. A. Hamaniuk , Bridging minds and machines: AI's role in enhancing mental health and productivity amidst Ukraine's challenges , in: S. O. Semerikov , A. M. Striuk , M. V. Marienko , O. P. Pinchuk (Eds.), Proceedings of the 7th International Workshop on Augmented Reality in Education (AREdu 2024 ), Kryvyi Rih, Ukraine, May 14 , 2024 , volume 3918 of CEUR Workshop Proceedings, CEUR-WS.org , 2024 , pp. 43 - 59 . URL: https://ceur-ws. org/ Vol- 3918 /paper020.pdf.

[17] O. V. Talaver , T. A. Vakaliuk , A model for improving the accuracy of educational content created by generative AI , in: S. O. Semerikov , A. M. Striuk , M. V. Marienko , O. P. Pinchuk (Eds.), Proceedings of the 7th International Workshop on Augmented Reality in Education (AREdu 2024 ), Kryvyi Rih, Ukraine, May 14 , 2024 , volume 3918 of CEUR Workshop Proceedings, CEUR-WS.org , 2024 , pp. 149 - 158 . URL: https://ceur-ws. org/ Vol- 3918 /paper372.pdf.

[18] W. Y. Leong , J. B. Zhang , AI on Academic Integrity and Plagiarism Detection, ASM Science Journal 20 ( 2025 ). doi: 10 .32802/ASMSCJ. 2025 . 1918 .

[19] B. Fajt , E. Schiller, ChatGPT in Academia: University Students' Attitudes Towards the use of ChatGPT and Plagiarism , Journal of Academic Ethics 23 ( 2025 ) 1363 - 1382 . doi: 10 .1007/ s10805-025-09603-5.

[20] P. Baron , Are AI Detection and Plagiarism Similarity Scores Worthwhile in the Age of ChatGPT and Other Generative AI?, SOTL in the South 8 ( 2024 ) 151 - 179 . doi: 10 .36615/sotls.v8i2. 411 .

[21] J. Sherice , 7 Common Types Of Plagiarism - Examples And How To Avoid, 2024 . URL: https: //originality.ai/blog/common -types-of-plagiarism.

[22] O. Sharyhin , O. Klochko , I. Tverdokhlib , Automation of plagiarism and ai detection in it students papers with a software tool using api , in: Proceedings of the 13th International Conference on Applied Innovations in IT (ICAIIT) , April 2025 , 2025 , pp. 269 - 277 . doi: 10 .25673/119243.

[23] O. A. Sharyhin , O. V. Klochko , Automation of checking student assignments in IT-related subjects based on AI systems , in: S. Papadakis (Ed.), Proceedings of the IX International Workshop on Professional Retraining and Life-Long Learning using ICT: Person-oriented Approach (3L-Person 2024) co-located with 19th International Conference on ICT in Education, Research, and Industrial Applications (ICTERI 2024 ), Lviv, Ukraine, September 23 , 2024 , volume 3781 of CEUR Workshop Proceedings, CEUR-WS.org , 2024 , pp. 88 - 97 . URL: https://ceur-ws. org/ Vol- 3781 /paper07.pdf.

[24] I. Festas , A. Seixas , A. Matos , Plagiarism as an Academic Literacy Issue: The Comprehension, Writing and Consulting Strategies of Portuguese University Students, International Journal for Educational Integrity 18 ( 2022 ) 25 . doi: 10 .1007/s40979-022-00119-8.

[25] Paraphraser .io, Paraphraser plagiarism checker, 2025 . URL: https://www.paraphraser.io/ plagiarism-checker.

[26] Zucol Services PVT LTD , Desklib: Detect plagiarism instantly with AI-powered precision , 2025 . URL: https://desklib.com/plagiarism-checker/.

[27] JustDone , Plagiarism Checker by JustDone, 2025 . URL: https://justdone.com/plagiarism-checker.

[28] 1text.com, 1text.com Plagiarism Checker , 2025 . URL: https://1text.com/plagiarismchecker.

[29] AppStar Inc , ExpertChat AI: Smart Plagiarism Checker , 2025 . URL: https://tutorial.expertchat.me/ ai-master/plagiarism.