-Learning to program is crucial in computer science degree programmes. For students to gain this skill, they need to practise a lot, since programming is a difficult and complex process and practice improves it. The courses "Algorithmics" and "Introduction to programming" have therefore become almost insurmountable barriers with high failure rates. To overcome these barriers, researchers have sought to find new ways of teaching. For their part, students have looked for ways to succeed with less effort by repeatedly resorting to plagiarism. The need for classes to be taught remotely due to the pandemic has further aggravated this problem. To fight it, we created the Algorithmi tool and equipped it with a plagiarism detection module. This tool promotes stand-alone study by allowing self-correction of the exercises whilst validating authorship. This paper presents a case study on the application of the plagiarism module to the exercise resolutions submitted by the students in their personal repository. Through the analysis of the results obtained we can conclude that there was a high rate of plagiarism. The current version of the tool needs to be extended with new features, not only to become more accurate, but above all to prevent plagiarism from happening.
INTRODUCTION
Programming is the art of making computers to solve
problems automatically. For this to happen the programmer
needs to have knowledge of problem-solving and
computational logic and master the programming language. A
task which is “considered difficult, complex and categorised
as part of the seven grand challenges in computing education”
[
This scenario has led many researchers to look for new
methodologies, strategies and tools to overcome the problem
[
Simultaneously, students also try to find other ways to succeed in the module, including plagiarism.
Several studies indicate that academic misconduct, and in
particular plagiarism, is particularly problematic in the
courses of programming [
Plagiarism is defined as the action of copying someone else's work or ideas without giving them credit for it. The existence of the Internet with original contents indexed, searchable and cloneable makes plagiarism accessible at the distance of a click. On the other hand, the existence of equipment and instantaneous communication applications such as smartphones facilitates the plagiarism of exercises with the transmission of answers to questions between students in an irregular way. In a regular classroom context, these factors can be eliminated or at least diminished through the presence and control of the lecturers. When their presence is not possible (as was the case during the confinement induced by the COVID-19 outbreak), these factors contribute to the increase of plagiarised works from the original sources and among the students.
Intensive practice is the best way for students to learn how to program correctly because it improves abstract thinking for complex problem solving. The need to perform many exercises in conjunction with overcrowded classes means that validation of the authorship of the exercises cannot be done manually by the lecturers in charge. This makes plagiarism the fastest and easiest way for students to succeed.
The pandemic outbreak caused by COVID-19 and the need for classes to be held remotely has further aggravated this situation, leading to the need for a tool that allows stand-alone coursework and detects plagiarism. It is in this scenario that the Algorithmi's plagiarism detection module appears.
This paper presents a case study of the use of the Algorithmi tool in students' answers to programming exercises in the course “Introduction to Programming” of the school year 2019/2020.
We start by addressing the issue of plagiarism, and then explain how Algorithmi detects plagiarism. The case study and the research conclusions are described below.
II.
PLAGIARISM
According to [
There are several implications regarding the practice of
plagiarism, some of these implications in the academic field,
according to [
The project Impact of Policies for Plagiarism in Higher
Education Across Europe (IPPHEAE), which took place
from October 2010 to September 2013, makes a comparison
of policies for academic integrity in Higher Education in the
European Union [
The scenario where students and faculty have no
knowledge about conduct that may constitute plagiarism or
fraud is worrying, because in addition to the legal
implications, plagiarism can also affect student learning,
considering that the simple fact of "copying and pasting"
inhibits critical thinking and the development of logical
reasoning. According to [
In addition, [
The most common methods of plagiarism according
to [
A. Plagiarism Detection in Source Code
As already mentioned, plagiarism does not occur only in
textual documents, or image files, but also in source codes.
According to [
The possible reasons for the practice of plagiarism as
described by [
There are several approaches in the literature regarding the detection of plagiarism, varying in implementation style and source code. Above all, it is important to emphasise initially the difference between detecting plagiarism in text files and in files derived from codes and algorithms.
According to [
Text file detectors are not effective for verifying
similarities between source files because this approach
cannot verify changes made to the copied code such as
changing variables and possible changes to the structure of
the document [
This modelling process usually follows some pre-defined steps that are established based on each objective and need. These factors may vary depending on the wide variety of plagiarism detectors available on the market. The most common practices for plagiarising source code files imply changing the code. Such practices range from changing the name of variables to inserting redundant code snippets. Figure 1 illustrates some of the most common techniques in source code plagiarism. In literature, these practices are characterised by "blinding methods", where the plagiarist aims to hide code snippets that are originally from another person.
Determining the level of similarity between two source
codes is a difficult task and several automatic tools have been
developed to help lecturers in the arduous task of detecting
cases of plagiarism. Examples of these software tools are
MOSS (Measure of Software Similarity) [
III.
PLAGIARISM DETECTION IN ALGORITHMI
Algorithmi is an information system to support learning programming consisting of a multi-platform application (IDE) designed to help students take their first steps in programming and a server which provides various services that can be accessed directly from a web browser, Figure 2.
The IDE not only shows and allows editing the flowchart and pseudocode, but also allows the display in several programming languages (Python, Java, C, C++, C#, PHP and JavaScript), Figure 4.
The server component is available through registration
and authentication in the same domain. The server used was
implemented with open-source technologies Apache,
MySQL and PHP that allow the provision of various features
to the information system. This server provides services that
access a common database: Backoffice - a web application
used by faculty, Frontoffice - a web application where
students can find solved problems and their score, and an API
- a set of REST web services ttha includes plagiarism
checker.
The application is available for free at
www.algorithmi.ipt.pt, and allows students to encode the
algorithms using flowcharts or pseudocodes. The application
internally uses a marker-based language, GAL (Generic
Algorithm Language) [
The GAL language supports the definition of simple and indexed variables of several types: integer, real, logical and text; it has I/O console instructions; it allows to control the execution flow through decisions and iterations and supports the definition of subalgorithms which can be recursive. This set of instructions allows coding complex algorithms through a set of tools designed for teaching and learning algorithms.
The algorithm can be edited in both languages and its visualisation can be displayed simultaneously as shown in Figure 3.
The backoffice component is used by faculty to create and make available programming exercises that can be accessed through the IDE, provided that the student is authenticated in the information system. The programming exercises can be corrected automatically and, in case of errors, the IDE provides a description of the errors made by students. This feature is particularly important in order to promote autonomous learning by students, as the IDE helps students verify that their algorithms are correct, and provides clues for their resolution.
After solving the exercises, the students submit them to the server that stores them in the student's personal repository. The student can visualise the resolution of their exercises in a web environment using the same tools as the IDE: flowchart, pseudocodes or programming languages; and can also execute them by translating them into Javascript embedded in a web page.
Figure 3 - Simultaneous editing and display of an algorithm. a) b)
Figure 5 - Programming exercise: a) Problem Statement; b) Automatic correction of the exercise by the
IDE.
A. Plagiarism control in Algorithmi
Algorithmi encourages stand-alone programming through
the exercise assessment tool and the tutorial system that
identifies errors. Problem repository is one of the forms of
assessment that can be used to assign a score to students, and
the provision of a roll of honour listing the names of the
students with the highest scores. Gamification is a key
component in learning environments to maintain students’
motivation to solve more complex exercises [
The pressure to get a good score in the repository and the opportunity to belong to the roll of honour make students plagiarise some coursework. The IDE provides some tools that limit source code plagiarism. The exercises provided by the server to the IDE are marked with the username which prevents other students from taking them as their own. The exercise delivery and assessment module is only available for authenticated users and the programs of that user. The other tool that prevents plagiarism is the cut and paste tool. Pasting from software programmes within and outside Algorithmi are only allowed within software programmes of the same user. This way we prevent the source code from being shared between users.
With the help of these two features we can ensure that the algorithm was built by the authenticated user, but we can not ensure that it was not copied from another user.
In order to inform the lecturer of this type of plagiarism, we introduced the plagiarism detection module in the information system to compare the similarity between the resolutions of the students submitted in their repository. The module was developed using JPlag (sources available at https://github.com/jplag/jplag) which communicates with the information system through a web service. Due to the Algorithmi's ability to convert the algorithm into various programming languages, we use the Java language to compare source codes and detect plagiarism.
The service receives as passing parameter two programmes in Java and returns a number in the interval [0.100] that represents the plagiarism prevalence rate. The results are presented through the backoffice with the results of the plagiarism prevalence rate for each programming exercise and with aggregate information for each worksheet and for each student.
CASE STUDY
This section analyses the application of the plagiarism module to the exercises submitted in the course “Introduction to Programming” of the CteSP programme in Information Systems Technologies and Programming offered at the Escola Superior de Tecnologia de Tomar - Instituto Politécnico de Tomar. The course covers the basic programming concepts: variables, operators, functions and control flow. One of the assessment components of the course is made based on the portfolio of programming exercises submitted by students. The exercises are divided into 15 worksheets with a total of 480 different exercises which cover the various subject contents. In each worksheet, the exercises have difficulty levels: o Demonstration - Solved exercises and problems that are provided to students to demonstrate a new concept. These exercises are crucial for stand-alone learning and students can edit them in order to check the outcome of these changes. o Easy - Exercises that are easy to solve and are usually variants of demonstration exercises. o Normal - Medium-difficulty exercises o Difficult - High-difficulty exercises where algorithmic techniques are explored, especially in the limit cases. o Challenging - Challenging exercises whose resolution depends on the combination of several complex algorithmic techniques
The greater the difficulty of resolution, the more points the student earns, improving his ongoing assessment component and consequent rise in the roll of honour position. Plagiarism in demonstration exercises is almost 100%. The lecturer-in-charge provides the resolution, the students can then execute and submit it and consequently the resolutions.
Figure 6 gives the plagiarism prevalence rate of an easy exercise solved by 18 identified students from S01 to S018, and whose plagiarism indexes were provided by the plagiarism detection module. The numerical information is presented in tabular form with the plagiarism prevalence rate between each of the submissions for the same exercise, together with a colour that ranges between green and red for easier reading of the results. By the empirical analysis of the preliminary results we consider that two algorithms are plagiarised if their similarity is equal to or higher than 90%. This information is presented in the last row and in the last column of Figure 6 and is useful to verify the degree of originality of the resolution of the exercise. In Figure 6 we can see that there are 10 resolutions that the system identifies as plagiarised and one that differs from all the others. This plagiarism prevalence rate is natural, since it is an easy exercise based on a similar demonstration problem. exercise. Due to the complexity of the algorithm that solves the exercise, the cases identified as plagiarism will be really plagiarism. In that figure we can identify a set of students who have a degree of similarity of 100% and a set of unique resolutions.
The previous figures present various information about the rate of plagiarism of worksheet exercises. The course has a strong component of stand-alone work and the great majority of the exercises are completed outside the contact hours, however the students continue to maintain contact with each other. Since Algorithmi does not allow you to make full copies of exercises, even if students exchange solutions with each other, they will have to edit the algorithms themselves. For these reasons, the plagiarism rates of the exercises in the worksheets are high. assessed practical test. Assessed practical tests are done individually during the contact hours and in the presence of the lecturer. The class is divided in shifts to increase the physical distance between students and the time for solving the exercises is limited. From the results presented in Figure 8 we can see that the plagiarism prevalence rate has decreased, however we can see that there are at least three resolutions that the system has identified as 100% plagiarism. Alunos S01 S02 S03 S04 S05 S06 S07 S08 S09 S10 >=90% S01 S02 S03 S04 S05 S06 S07 S08 S09 S10 >=90% 2 100 51 70 60 51 52 84 51 100 41 51 100 51 46 100 81 49 60 51 61 2 100 70 51 60 51 52 84 51 100 41 2 100 60 46 60 46 47 58 46 60 53 1 51 100 100 51 46 81 49 51 61 3 100 100 52 81 52 47 81 50 81 52 63 1 84 49 84 58 49 50 49 84 39 1 100 51 60 51 46 81 49 51 61 2 100 100 100 51 100 60 51 52 84 51 41 3 100 41 61 41 53 61 63 39 61 41 100 1 2 2 2 1 3 1 1 2 3 1 1,8
Student portfolios consists of more than 7,000 exercises evaluated by the Algorithmi module to obtain a grade and by the plagiarism module to obtain the similarity index. Figure 9 gives an overview of average plagiarism in each student's exercises and the percentage of exercises solved. The overwhelming majority of students have solved more than 50% of the proposed exercises and more than half have solved more than 90%. This statistic shows the work and commitment of the students in solving the exercises and because "you only learn to program by programming" the result was reflected in the score of the repository.
Another important conclusion is that the plagiarism prevalence rate varies according to the number of exercises solved by the students. Students with fewer solved exercises tend to plagiarise more, due to their learning difficulties. Since Algorithmi doesn't let you make direct copies and deliver them as your own, plagiarising takes work and students replace the work of thinking and creating with the work of copyists that is far less rewarding.
Average levels of plagiarism have always remained high, in the order of 80%, which suggests a high rate of plagiarism. was designed to increase the effectiveness of this system for teaching/learning computer programming.
This module was applied to the course which was taught in the first semester of the school year 2019/2020 to check the rates of plagiarism in the students repository.
It was found that the rates of plagiarism are high, especially in the worksheets completed outside contact hours. In assessed practical exercises, whose objective is to assess knowledge, the plagiarism rates are significantly lower. This is because the exercises are more difficult, the students have a limited time for their resolution and were solved during contact hours. Nevertheless, the system detected several cases of plagiarism.
With the COVID-19 forced confinement of students and staff, in which the majority of teaching was delivered remotely, the module now developed has assumed greater significance. The work developed corresponded to expectations and above all opened new horizons that will be explored in future versions.
In the future, we intend to apply the online module where the system registers and informs the student about the originality of the submitted code. In the offline version, whose results we present here, we only know which exercises are plagiarised, but we do not know the original source. With the online version, the first exercise to be submitted has a maximum degree of originality while copies submitted later will see their originality diminished. This way we intend to stimulate the development of different algorithms for the same problem, increasing the gamification already present in Algorithmi.
The fact that students realise that the system has a system that detects plagiarism may serve as a deterrent for them to avoid less ethical behaviour and channel their energy to the development of algorithmic reasoning, because in Algorithmi plagiarism also takes work.