The paper presents a new approach to Arabic Sentiment Analysis (ASA) using Bi-Directional Long Short-Term Memory (BiLSTM) and Convolutional Neural Networks (CNN)-BiLSTM models. The proposed framework introduces a noise layer to these models to overcome the issue of overfitting, which is common in ASA. The framework also includes a local surrogate explainable model to provide explanations for specific predictions, making it more transparent. The researchers conducted experiments on public benchmark Arabic SA datasets and found that the inclusion of noise layers improved performance in sentiment analysis for the Arabic language. The proposed method outperformed some known state-of-the-art methods.
Publication date: 24 Sep 2023
Project Page: https://arxiv.org/abs/2309.13731v1
Paper: https://arxiv.org/pdf/2309.13731
