Author Identification Using LSTM and BiLSTM Techniques Applied on Imbalanced Data Set with Hyper Encoding Parameters

This work explores a series of authorship attribution approaches based on deep neural networks. Authorship attribution is defined as the identification of the author of a piece of text. It has attracted the attention of many researchers due to its relevance to a wide range of applications, including forensic investigations and plagiarism detection. A large number of features and approaches have been used for this task. However, there has been a lack of studies that involve multiple datasets or the use of a variety of deep learning based classifiers. In this research work, we explore several deep neural network models in author based text classification experiments. Then, we compare these approaches in order to find the best authorship attribution method. We present the concepts and techniques of deep learning in author attribution which are applied mainly to literary corpora and based on features defined to quantify the style of an author such as sentence and word length, word frequency and vocabulary richness. The main used models are: Flow Control Recurrent Networks (LSTM). This work allowed us to explore some deep learning techniques such as LSTM, Bi-LSTM. We obtained satisfactory results on our dataset using Glove vectors by reaching a maximum score of 94% with Bi-LSTM and 93% for LSTM. The results obtained in this investigation are motivating and suggest to explore much larger corpora for more significance.