Semantic Watermarks for Detecting Cheating in Online Database Exams

Abstract

Due to the COVID-19 pandemic,we were forced to conduct the exam for a database course as an online exam.An essential part of the exam was to write non-trivial SQL queries for given tasks.In order to ensure that cheating has a certain risk,we used several techniques to detect cases of plagiarism.One technique was to use a kind of ``watermarks'' invariants of the exercises that are randomly assigned to the students.Each variant is marked by small variationsthat need to be included in submitted solutions.Those markers might go through undetectedwhen a student decides to copy a solution from someone else.In this case,the student would reveal to know a ``secret''that he cannot know without the forbidden communication with another student.This can be used as a proof for plagiarisminstead of just a subjective feeling about the likelihoodof similar solutions without communication.We also used a log of SQL queries that were tried during the exam.Finally,we evaluated similarity-based techniques for SQL plagiarism detection.