Choosing Code Segments to Exclude from Code Similarity Detection

Abstract

When student programs are compared for similarity as a step in the detection of academic misconduct, certain segments of code are always sure to be similar but are no cause for suspicion. Some of these segments are boilerplate code (e.g. public static void main String [] args) and some will be code that was provided to students as part of the assessment specification. This working group explores these and other types of code that are legitimately common in student assessments and can therefore be excluded from similarity checking. From their own institutions, working group members collected assessment submissions that together encompass a wide variety of assessment tasks in a wide variety of programming languages. The submissions were analysed to determine what sorts of code segment arose frequently in each assessment task. The group has found that common code can arise in programming assessment tasks when it is required for compilation purposes; when it reflects an intuitive way to undertake part or all of the task in question; when it can be legitimately copied from external sources; and when it has been suggested by people with whom many of the students have been in contact. A further finding is that the nature and size of the common code fragments vary with course level and with task complexity. An informal survey of programming educators confirms the group's findings and gives some reasons why various educators include code when setting programming assignments.