Classification of Arabic Social Media Texts Based on a Deep Learning Multi-Tasks Model

Abstract

The proliferation of social networking sites and their user base has led to an exponential increase in the amount of data generated on a daily basis. Textual content is one type of data that is commonly found on these platforms, and it has been shown to have a significant impact on decision-making processes at the individual, group, and national levels. One of the most important and largest part of this data are the texts that express human intentions, feelings and condition. Understanding these texts is one of the biggest challenges that facing data analysis. It is the backbone for understanding people, their orientations, and making decisions in many cases and thus predicting their behavior. In this paper, a model was proposed for understanding texts that written by people on social media platforms, and hence knowing people's attitudes within specific topics, the emotion of those people, positivity, negativity, and neutrality. Also, it extracts emotion of those people. In this context, the system solves many tasks in natural language processing therefore it uses many techniques including topic classifier, sentiment analyzer, sarcasm detector and emotion classifier. CNN-BiLSTM was used for topic classifier, sentiment analyzer, sarcasm detector, and emotion classifier where (f-measure, accuracy) were (97,97.58) %, (84,86) %, (95,97) %, and (82,81.6) % respectively.