Twitter Opinion Leaders’ Detection (TOLD): A comprehensive approach
Malek Chebil, Mohamed Anis Bach Tobji, Rim Jallouli · Procedia Computer Science · 2024
Opinion leaders are pivotal in shaping public opinion, influencing marketing strategies, guiding administrative decision-making and driving political discussions in online social networks. However, developing a standardized algorithm for detecting opinion leaders across diverse platforms is challenging due to the unique characteristics of different data sources. Each source has its structural properties, necessitating a tailored detection algorithm. In this paper, we propose a new hybrid approach, Twitter Opinion Leaders’ Detection (TOLD), designed for identifying opinion leaders on Twitter. TOLD combines a machine learning approach for computing user score in topical influence and domain influence, a statistical approach for measuring engagement and activity user score, a graph-based approach for measuring network structure influence user score and a PageRank-based approach for measuring popularity influence user score. We then identify opinion leaders through these multidimensional scores in each domain and across all domains. We compare TOLD with Topic-sensitive PageRank and TwitterRank methods, assessing their correlation and similarity. The results show that TOLD outperforms the other methods. Our approach offers a comprehensive and precise detection of opinion leaders related to particular domains, enriching our understanding of influential individuals in online communities.