ABSTRACT

Social media has led to the creation of complex networks of political and social leaders, and has allowed for instantaneous interaction with broad audiences. Semantic analysis of these interactions is essential for policy-makers and various experts, such as in the anti-terrorism community. However, comprehensive quantitative analysis of the attitudes of Arab political and religious leaders is currently limited. To handle this problem, we first create a novel dataset consisting of 1,145,525 tweets posted in 2009–2018 from 449 political, social, and religious leaders from 12 Arab countries. Next, to semantically process various Arabic dialects in their informal expressions on social media, we develop a Latent Semantic Analysis algorithm that is then used to identify the individuals with the highest semantic matches to our pre-specified categories of interest. Finally, we construct an original Temporal Semantic Similarity (TSS) measure that allows the tracking of those expressions over time. Experimental results demonstrate that the proposed semantic approach can effectively and efficiently process documents in each category. We have made all the data, code and results publicly available on the project homepage: https://github.com/ArabLeader.