JPlag DOs and DONTs

[sebinside]

JPlag is - without a doubt - a powerful tool to counter software plagiarism. However, it is not designed to be used autonomously without human supervision. Recently, this raised ethical concerns. To face these, we want to create a list of DOs and DONTs, and provide information on when not to use JPlag. This shall help non-expert users to better understand what JPlag can and can't do.

Everyone is invited to add to this list which will later be persisted in #267. Some examples:

DOs

• Use JPlag on large tasks and submissions with little general matches, e.g., in algorithms
• Use the distribution to get familiar with the overall (potentially false positive) similarity among all submissions
• In case of presumed plagiarism, double-check the affected source could by hand

DON'Ts

• Do not use JPlag on small tasks, e.g., algorithms that will be implemented very similar
• Do not define strict thresholds when a submission is flagged as plagiarism
• Never use JPlag without human supervision of the results
• Don't assume that the code will be compared to the internet

[tsaglam]

We should also clarify that we consider the text/char frontends as severely outdated, as more modern (ML/NLP-based) methods might be way better to detect plagiarism in natural language texts. Thus, these frontends should be used with care.

[sebinside]

In the same train of thought, we should rework the README description in a way that implies that JPlag only helps in plagiarism detection and is no standalone detector without the help of the human.