Sentiment Analysis of Arabic Tweets using ARABERT as a fine tuner and feature extractors
Athir Mohammed Alsugair, Norah Saleh Alghamdi · 2024
The influence of social media platforms on our daily lives is significant, and Twitter is one such platform that can serve as a valuable source for gathering public opinion on various products, services, and events. Sentiment analysis is a technique that involves analysing the emotions and attitudes expressed by the public toward specific topics, which can be categorized as positive, negative, or neutral. The aim of this study is to provide the performance of sentiment analysis for Arabic tweets related to airline services in Saudi Arabia using two approaches: fine-tuning AraBERT and using AraBERT as a feature extractor. AraBERT is a pre-trained language model that can generate contextualized word embeddings, which capture the meaning of words in the context of the surrounding text. Our corpus contained 9K airline tweets in Arabic. We will evaluate the two approaches' performance based on their accuracy, precision, recall, and F1 score. The outcome of this project shows that fine-tuning AraBERT performs better than using AraBERT as a feature extractor for sentiment analysis of Arabic tweets related to airline services in Saudi Arabia. The fine-tuned AraBERT model achieved an accuracy rate of 88%, whereas the AraBERT feature extractor model achieved an accuracy rate of 70%.