Rumor Detection in Algerian Arabizi Based on Deep Learning and Associations

Abstract

AbstractInfobesity has become a reality these days and we are inundated with news via social media. This rapid large-scale transmission is not always synonymous with credibility of information, which creates a thriving ecosystem for the spread of rumors. The identification of rumors on social networks is one of the most recent and important issues, because of its impacts and the difficulty of manual processing.In this work, we propose a rumor detection approach in the Algerian arabizi. It extracts information with the associations between rumors and reactions of social network users. The features extracted from the associations and other features representing the semantics of the users’ expression, are feed to an attention mechanism to compute their importance used in the construction of deep learning models. Different classification models (LSTM, GRU and CNN) and textual representations (Word2vec, bag of ngrams and ELMo) were used to test their suitability for the study of associations between the rumors and the reactions of social network users.KeywordsRumor detectionText classificationDeep learningAssociations