Kato - A Plagiarism Detection Tool for DLV

Abstract

Kato is a plagiarism detection tool for DLV-programs. It was developed to detect similar programs in collections of student-assignment solutions from logic programming courses at the Vienna University of Technology. Kato provides four different tests for calculating similarity values between two files: two string-based tests, one attribute-based test, and one structure test. Each of them yields a similarity value between 0 (no similarity) and 1 (perfect match). Currently, the tool is realised for DLV programs but it is designed to handle further ASP solver dialects as well.

Installation

Kato, entirely developed in Java (version 6.0), is available as Java archive. Since Kato is used for plagiarism detection in current lectures, we can provide the entire Kato Java archive only for researchers and lecturers on request.

Requirements

• Java Runtime Environment 6.0 (or higher)

Optional

• MySQL (http://www.mysql.com/)

Depending on the number of files to test, the size of the files, and the applied tests, Kato may need more memory than the Java VM normally provides. Otherwise, 'OutOfMemoryErrors' may appear. If this happens, the memory for the Java VM can be increased by starting Kato with the Java parameters ``-Xmx512M -Xss64M'', where Xmx is the maximum available heap size (512MB) and Xss sets the thread stack size (64MB).

Launching Kato

java -jar kato.jar

or

java -Xmx512M -Xss64M -jar kato.jar

Description

Kato is a plagiarism detection tool for finding similar pairs of programs stemming from student-assignment solutions and thus can perform pairwise similarity tests on rather large collections of programs. Currently, Kato can handle the following covering strategies of plagiarists:

• reordering rules,
• reordering predicates in rules,
• reordering predicates in rules,
• reordering predicates in rules,
• different syntax styles of build-in predicates, and
• changing the syntactic form of build-in predicates.

Following a hybrid approach, Kato performs four kinds of comparison tests to find suspected copies. Each test yields a similarity value between 0 (no similarity) and 1 (perfect match).

• String-based comment tests. When programs are copied, it is surprisingly quite frequent that program comments are copied as well and usually little effort is spent to mask such copied comments. Consequently, this test aims at revealing similar parts in program comments.
• String-based program tests. This test is very similar to the comment test but considers entire programs. Thus, this test compares the cleansed source code of two programs interpreted as strings.
• Fingerprint tests. This tests compares the fingerprints of programs pairwise, where a fingerprint is a collection of relevant statistical program attributes like hash-codes, the number of rules, the number of predicates, the number of constants, program size, and so on.
• Structure tests. This test operates on the program representation that was built in a preceding parsing step. The notion of program similarity that underlies the structure test is designed to thwart simple disguising strategies like permuting rules or literals as well as more advanced camouflage techniques like uniformly renaming variables in rules or renaming auxiliary predicates in programs. To make the similarity function sensitive to common rule patterns, we also implemented a context dependent notion of confidence regarding plagiarism: When whole collections of programs are examined for suspicious pairs of programs, a global occurrence table for rules gives additional information on how specific two rules are. This information is then used to compute a confidence measure with respect to plagiarism.

The system was entirely developed in Java (version 6.0) and provides a graphical user interface for controlling test runs, where a test run is a series of pairwise program comparisons on collections of programs or single program pairs. The user can modify the applied tests and configure how the system displays results. The results of a test run can be saved in a file or exported into a database---currently, only MySQL databases are supported, however.Test results are displayed in a table where each line gives the results of various pairwise program comparisons. For results of single tests, Kato provides more detailed views with details regarding the computed similarities.

Feedback

E-mail to: {oetsch,puehrer,schwengerer,tompits} "at" kr.tuwien.ac.at

Last update: 09.02.2010