Computational methods to detect plagiarism in assessment

While many institutions of higher education offer courses via distance education, there is one aspect which is difficult to realise by use of the Internet only: assessment. If exams are performed online, how can the course provider guarantee that the student participating in the exam is the person enrolled? Without any Internet-based form of authenticating the student's identity, flexible delivery can break down at this point. As a consequence, traditional identity checks are introduced such as requiring the student to be physically present and to take the exam at a local institution, or requiring the student to sign documents that certify his/her identity. This paper discusses assessment in flexible delivery and how plagiarism can be detected. It presents a method for testing the identity of a student (or more generally, author) online, without any interference with the examination process. Recent advances in computational text analysis allow authorship identification with high reliability. That is, the original author of a document submitted for assessment can be determined successfully with an accuracy and precision of well above 90 percent. The computational methods include machine learning techniques such as "support vector machines", which are highly successful in text classification and a range of other practical applications