Enhancing Contextualised Language Models with Static Character and Word Embeddings for Emotional Intensity and Sentiment Strength Detection in Arabic Tweets

Many studies have focused on Arabic sentiment or emotion classification tasks. However, research alternative aspects of affect, such as emotional intensity and strength tasks, has been somewhat limited. In this paper, we propose a method for enriching contextualised language model that incorporates static character word embeddings in tweets. We examine the assumption models using are trained specifically corpus containing extensive affect-related words can boost performance models. Through development character-level embeddings, found our is able to overcome limitations associated with out-of-vocabulary words, which common problem when dealing informal text. Given this, developed achieves state-of-the-art results detection social media.