Author Attribution with CNN’s

In this report, the results from my CS224D final project are given and explained. The project was based on the application of relatively new neural network architectures, namely convolutional neural networks over word embeddings, to the task of authorship identification. The problem was posed as a classification task and models were evaluated over two datasets, a baseline of my own collection and a competition dataset with pre-existing, published results. Despite robust results on the baseline dataset, the CNN architecture failed to outmatch the best competition submission, although it did outscore most of the competitors’ submissions.