Measuring User Influence in Real-Time on Twitter Using Behavioural Features

Identifying influential users in online social networks, specifically Twitter, has drawn considerable research interest due to their profound influence on real-world events. Existing studies have primarily focused on topological aspects to identify influential users, which necessitate a complete network structure and ignore the behavioural aspects of the users. However, real-world network structure can be challenging to obtain. Thus, this study presents the BehaviourRank Influence (BRI) method, which focuses on evaluating Twitter user influence by analysing behavioural characteristics instead of traditional network structures. This approach acquires data in real-time to derive six behavioural characteristics by employing new metrics to achieve accurate and uniform characterizations. The proposed metrics are compared to baseline metrics from the literature using a user classification prediction model to validate their predictive ability. It also ranked behavioural characteristics by importance using SHAP and Information Gain approaches. To determine overall influence, a unique Rank-based Exponential Smoothing algorithm is proposed. The findings indicate that most features incorporating proposed measures exhibited superior performance compared to features with baseline metrics across all three datasets. The proposed feature sets achieved accuracy scores of 86.67%, 73.47%, and 85.87% in the respective datasets, while the baselined feature sets attained scores of 64.23%, 63.30%, and 66.30%. Moreover, the Rank-Based influence measurements algorithm provides a detailed and adaptable structure for measuring influence, recognising the significance of feature ranks in capturing the evolving nature of user behaviour.