Maintaining Academic Integrity in Programming: Locality-Sensitive Hashing and Recommendations

Abstract

Not many efficient similarity detectors are employed in practice to maintain academic integrity. Perhaps it is because they lack intuitive reports for investigation, they only have a command line interface, and/or they are not publicly accessible. This paper presents SSTRANGE, an efficient similarity detector with locality-sensitive hashing (MinHash and Super-Bit). The tool features intuitive reports for investigation and a graphical user interface. Further, it is accessible on GitHub. SSTRANGE was evaluated on the SOCO dataset under two performance metrics: f-score and processing time. The evaluation shows that both MinHash and Super-Bit are more efficient than their predecessors (Cosine and Jaccard with 60% less processing time) and a common similarity measurement (running Karp-Rabin greedy string tiling with 99% less processing time). Further, the effectiveness trade-off is still reasonable (no more than 24%). Higher effectiveness can be obtained by tuning the number of clusters and stages. To encourage the use of automated similarity detectors, we provide ten recommendations for instructors interested in employing such detectors for the first time. These include consideration of assessment design, irregular patterns of similarity, multiple similarity measurements, and effectiveness–efficiency trade-off. The recommendations are based on our 2.5-year experience employing similarity detectors (SSTRANGE’s predecessors) in 13 course offerings with various assessment designs.