Arabic Sentiment Analysis Based on Salp Swarm Algorithm with S-shaped Transfer Functions

Social media has played a significant role in marketing and advertising. Monitoring the attitude of customers and analyzing their written sentiments to evaluate their opinions toward a particular product, topic or situation becomes essential to improve the product quality and customer services. Due to the importance of sentiment analysis (SA), a plethora of tools and systems have been developed for analyzing the polarities of people’s sentiments. However, SA is a difficult task, especially when dealing with massive data resources. Feature selection (FS) algorithms are needed for the machine learning (ML) process to reduce the high dimensionality space. In this paper, we propose an enhancement of a bio-inspired optimizer, called the salp swarm algorithm (SSA) designed for feature selection to solve the problem of Arabic sentiment analysis. Our proposed algorithm operates in two phases: The first phase employs a filtering technique based on the information gain (IG) metric to reduce the number of features. The second phase employs a wrapper technique which combines the basic SSA optimizer with four variants of S-shaped transfer functions. Experimental results show that the SSA combined with the S-shaped transfer functions outperformed the particle swarms optimizer (PSO) and the grey wolf optimizer (GWO) in term of classification accuracy.