Salp Swarm Algorithm-Based Sentiment Analysis of Forest and Land Fire Data with Z-Shaped Transfer Functions

This paper propossed an advanced method to enhance sentiment analysis (SA) for large-scale data, such as the significant number of tweets generated during the forest and land fires in Indonesia in 2019, which became a trending topic with over 316,782 tweets. Due to the complexity of sentiment analysis, especially with vast datasets rich in features, the paper highlights the necessity of feature selection (FS) algorithms to effectively manage the space of high dimensions in data that cannot be handled with conventional machine learning. To address this challenge, we propose improving a bio-inspired optimization technique referred to as the salp swarm algorithm (SSA) for feature selection to enhance the sentiment analysis of tweets in Indonesian. The next challenge arose when it comes to the feature selection process, SSA typically converge slowly, especially in binary problems and large data. Our approach using a filtering method utilizing the wrapper method, integrating the SSA optimizer with four Z-shaped transfer function variants, which is applied to minimize feature quantity. Comparative experiments demonstrate that our SSA-enhanced method surpasses particle swarm optimizer (PSO), ant lion optimization (ALO), and conventional SSA regarding classification accuracy and time processing. The experimental findings revealed that the enhanced SSA integrated with Z4-shaped transfer functions achieved superior accuracy performance, reaching 89.95 %, utilizing a Support Vector Machine (SVM). However, this approach had an excessively long processing time compared to implementing a Z1-shaped transfer function with the Naive Bayes algorithm, which completed processing in just <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$2 \times 10^{-10}$</tex> seconds, with a not much different accuracy of 89.33 %.