Authorship Classification in a Resource Constraint Language Using Convolutional Neural Networks

Authorship classification is a method of automatically determining the appropriate author an unknown linguistic text. Although research on authorship has significantly progressed in high-resource languages, it at primitive stage realm resource-constraint languages like Bengali. This paper presents approach made Convolution Neural Networks (CNN) comprising four modules: embedding model generation, feature representation, classifier training and testing. For this purpose, work develops new corpus (named WEC) Bengali (called BACC-18), which are more robust terms authors’ classes unique words. Using three text techniques (Word2Vec, GloVe FastText) combinations different hyperparameters, 90 models created study. All assessed by intrinsic evaluators those selected 9 best performing out for classification. In total 36 models, including (CNN, LSTM, SVM, SGD) with 100, 200 250 dimensions, trained optimized hyperparameters tested benchmark datasets (BACC-18, BAAD16 LD). Among CNN achieved highest accuracies 93.45%, 95.02%, 98.67% BACC-18, BAAD16, LD, respectively.