Analysis of Arabic Tweet Sentiment About Trending Hashtags Using Transfer Learning and Machine Learning Models

This study used a multi-pronged approach to examine how Gulf region Twitter users felt about COVID-19 trending topics. The data collection phase of the study started with the selection of relevant hashtags and time frames for analysis. The Twitter API was then used to compile a representative dataset of Arabic tweets pertaining to the COVID-19 trending topics. Subsequently, the research proceeded to the data annotation stage, employing a hybrid annotation technique that fused the transfer learning model and lexicon-based approach to assign a sentiment label to every tweet. Analyzing the patterns of tweet distribution over time exposed interesting patterns and possible sentiment expression influencers. The research obtained good accuracy scores by using a sentiment analysis model that combined three popular machine learning algorithms (Multinomial Naive Bayes, CountVectorizer, and TfidfVectorizer) with three feature representations (Ngram, TfidfVectorizer, and CountVectorizer). The sentiment tendencies of Arabic-speaking Twitter users toward trending topics were revealed by these scores. With the Ngram(1,2) representation, the LinearSVC algorithm achieved an impressive accuracy score of 89.1%, making it stand out as the best performer among all feature representations.