Performance Comparison of Word2Vec Models for Detecting Arabic Hate Speech on Social Networks
Samar Al-Saqqa, Arafat Awajan, BASSAM H. HAMMO · 2022
Detecting hate speech has become increasingly important for online communities. Despite emerging research to address the problem, more efforts are still needed to improve the performance of detection methods for the Arabic language. In this paper, we compare the performance of two Word2Vec models; the continuous bag of words (CBoW) and skip-gram using a dataset collected from multiple social media platforms; Facebook, Twitter, YouTube, and Instagram. We compared seven machine learning (ML) algorithms and measured their performance using different evaluation metrics. The CBoW model outperformed the skip-gram model in terms of accuracy. In addition, using a higher embedding dimensional value resulted in a better performance.