Arabic Sentiment Classification Using Convolutional Neural Network and Differential Evolution Algorithm.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Machine Learning

Background:

• Convolutional Neural Networks (CNNs) show promise in Arabic sentence classification.
• Developing effective CNNs for Arabic sentiment analysis is complex and time-intensive.

Purpose of the Study:

• To address the complexity of building CNNs for Arabic sentiment classification.
• To introduce an automated approach for optimizing CNN architecture and parameters.

Main Methods:

• Combining the differential evolution (DE) algorithm with CNNs.
• Utilizing DE to automatically search for optimal CNN parameters: convolution filter sizes, number of filters per convolution filter size (NFCS), fully connected (FC) layer neurons, initialization mode, and dropout rate.
• Investigating the impact of DE's mutation and crossover operators.

Main Results:

• The proposed DE-CNN framework was evaluated on five Arabic sentiment datasets.
• DE-CNN demonstrated superior accuracy compared to state-of-the-art algorithms.
• The DE-CNN approach proved to be less time-consuming in achieving optimal configurations.

Conclusions:

• DE-CNN offers an efficient and effective solution for Arabic sentiment classification.
• Automated optimization using DE significantly enhances CNN performance for this task.
• The framework provides a valuable advancement for sentiment analysis in Arabic language processing.