Fair Event Influence Analysis in Social Networks

<p dir="ltr">Social network platforms have become an important tool to analyse the influence spread of real-world events. Two core challenges remain in event influence analysis: how to model an event’s propagation within a social network and how to optimize the propagation to maximize user-event engagement. Existing information propagation models deliver a large volume of irrelevant messages to users with good information access, while rarely activating users who lack sufficient information access. This unequal information access exacerbates the disparity of influence received by individual users, resulting in a fairness issue. Existing influence maximization solutions focus on activating the maximal number of users for a piece of information, but they lack adaptability to event-related information in a streaming environment where the content or context of an event evolves over time. Overlooking this dynamic nature of events leads to inaccurate influence estimation for users, which further hinders the effective propagation of events. Furthermore, existing solutions lack explanations for their results. Both influence and fairness are evaluated by predefined score functions. However, it is difficult for audiences without prior knowledge of social influence analysis to comprehend the implications of such scores. Consequently, in this thesis, two main contributions are made to analyse events’ influence in social networks: maximizing fair event-aware influence in microblogs and visualizing the spread of event influence.</p><p dir="ltr">We propose a Fair Event-aware Influence Maximization (FEIM) framework, which allows the influence analysis system to coordinate message propagation in an event-sensitive and fair manner. Specifically, we first build an event-aware social graph that dynamically captures the probability of each user being activated by different event messages. Then, a novel Ensemble Activation Forest (EAF) model is developed to enable an equal number of messages sent to each receiver in the information diffusion process. Finally, we design a set of optimization strategies, including a Dijkstra-like Fast Reachability Calculation (DFRC) algorithm, a Greedy Reachability-based Seed Selection (GRSS) algorithm and a Fair Edge Selection (FES) algorithm to improve the effectiveness and efficiency of fair event propagation. Extensive experiments are conducted to prove the superiority of our FEIM framework.</p><p dir="ltr">In our second study, we examine three case studies to visualize the event propagation of FEIM and two state-of-the-art Influence Maximization (IM) methods. Each case follows a consistent procedure. First, for each method, we construct a propagation graph using shared raw data to illustrate how the method perceives and responds to event-related content. Second, we depict the message transmission along each edge of these graphs during event propagation as planar figures to clearly and intuitively showcase influence spread results and fairness issues. Specifically, in the case of Nepal Earthquake 2015, we construct propagation graphs using social messages about sub-events “food” and “blood”; in the case of Texas Flood 2015, we investigate the propagation paths of sub-events “usa” and “water”; in the case of World Cup 2014, we focus on sub-events “team” and “game”.</p>