Neural Network-Based Author Attribution for Assamese Language: Insights from RNN and GRU Architectures

The study presented in this paper highlights the performance and error analysis of three neural network models for author attribution task of assamese language documents. The task of author attribution has been explored very deliberately in some high privileged language like English. However, the work of exploring this in some low-resource languages like Hindi, Kannada, Urdu and Assamese have been next to negligible. In fact, the attempt made through this paper to study the effectiveness of machine learning models on the author attribution task with respect to Assamese language is a first of its kind. Out of all the models, GRU came out as the best performing model with an accuracy of 81%. This study contributes valuable insights into neural network applications for low-resource languages, advancing the field of regional NLP research.