Evaluation of Arabic Long Answers Using a Hybrid Deep Learning Neural Network Model

This paper uses a hybrid deep learning methodology comprising three parts. Initially, we train a Stacked BiLSTM model, a Bidirectional Long Short-Term Memory model that adds multiple dense layers. Subsequently, we implement a Stacked BiLSTM and Convolutional Neural Network (CNN) hybrid. Using Stacked BiLSTM models with multiple dense layers offers several compelling advantages. These models capture complex dependencies within sequential data, enhance representation learning, increase modelling capacity, mitigate the vanishing gradient problem, and improve overall model performance. Using a combination of Stacked BiLSTM and CNN provides effective text representation, captures contextual dependencies, improves semantic understanding, handles variable-length input robustly, and enhances generalisation and performance. Using a combination of a Stacked GRU and CNN model in NLP tasks offers benefits such as effective modelling of sequential dependencies, efficient text feature extraction, improved contextual understanding, robust handling of variable-length input, and enhanced performance in a wide range of NLP tasks. Stacked BILSTM with multiple dense layers), the accuracy result is 90%. In the second model (Stacked BILSTM—CNN with multiple dense layers), the accuracy result is 92%. Finally, in the third model (GRU—CNN with multiple dense layers), the accuracy result is 95%.