A real-time framework for opinion spam detection in Arabic social networks

Abstract

In today’s interconnected digital landscape, social media platforms serve as the primary avenue for global conversations, encompassing various topics and opinions. Opinion spam entails spreading misleading content masked as authentic opinions. The propagation of opinion spam poses a significant challenge, undermining the authenticity and trustworthiness of online interactions and disturbing the unrestricted exchange of ideas. One of the main challenges in spam detection is the rapid flow of spam content, which necessitates real-time detection mechanisms. Additionally, another important obstacle in detecting spam on Arabic social networks is the limited availability of labeled data. This paper proposes a framework for Real-Time Arabic Opinion Spam Detection (RTAOSD) that was developed to effectively detect opinion spam within Arabic social networks. This framework integrates advanced machine learning models, sentiment Analysis, and real-time processing techniques to achieve accurate and efficient detection of opinion spam. Furthermore, RTAOSD categorizes the non-spam content according to its relevance to topic of interest in to purify the content appear to social network users. Experimental evaluations conducted on real-world datasets demonstrate the effectiveness of RTAOSD in detecting opinion spam which leads to provide users with filtered content that match with their interest and overcome the problem of information overloading. The proposed framework achieved macro-F1 scores for spam detection ranging from 91% to 99% on different Arabic datasets surpassing previous work. While for topic relevance classification, RTAOSD achieved a macro-F1 of 86% for binary relevance and 78% for categorical relevance outperforming previous approaches used. The outcomes of this research is a real-time Arabic spam detector that accurately detects spam content and classifies non-spam text according to its relevance to topic . In addition to providing a visualization module for analyzing and reporting the characteristics of the filtered text.