Methods for Identifying Versioned and Plagiarised Documents

The widespread use of online publishing of text promotes storage of multiple versions of documents and mirroring of documents in multiple locations, and greatly simplifies the task of plagiarising the work of others. We evaluate two families of methods for searching a collection to find documents that are co-derivative, that is, are versions or plagiarisms of each other. The first, the ranking family, uses information retrieval techniques; extending this family, we propose the identity measure, which is specifically designed for identification of co-derivative documents. The second, the fingerprinting family, uses hashing to generate a compact document description, which can then be compared to the fingerprints of the documents in the collection. We introduce a new method for evaluating the effectiveness of these techniques, and demonstrate it in practice. Using experiments on two collections, we demonstrate that the identity measure and the best fingerprinting technique are both able to accurately identify co-derivative documents. However, for fingerprinting parameters must be carefully chosen, and even so the identity measure is clearly superior.