Enhancing Similarity-Based Authorship Verification using Corpus
| Authors | |
|---|---|
| Year of publication | 2017 |
| Type | Appeared in Conference without Proceedings |
| MU Faculty or unit | |
| Citation | |
| Description | Authorship verification problem can be defined as a task to determine whether two given texts were or were not written by an identical author. A similar task of the authorship attribution consists of choosing one author out of a predefined set of candidate authors as the most probable composer of a given document. The second task is usually transformed to a classification problem where the authors represent category names. In this respect the authorship verification corresponds to an open-class variant of authorship attribution. As the authorship attribution task (in the closed-class variant) can be solved with significantly higher accuracy, we suggest to transform the problem of authorship verification to be more similar to authorship attribution using two novel techniques: Ranking Distance and Corpus Ranking. The results indicate that the problem transformation and application of our optimizations increases the accuracy of authorship verification algorithms. All experiments were performed on Czech books, Slovak Internet news and English SMS messages, however proposed algorithms are document-type and language independent. |
| Related projects: |