Metody maximalizace vlivu v sociálních sítích

Abstract

This thesis focuses on maximizing the spread of influence in social networks. The main objective is to design and implement an application used for testing and conducting experiments to compare diffusion models and algorithms for selecting an initial set of nodes. The developed web application, built using Python (FastAPI) and React technologies, allows users to run simulations using the Independent Cascade, Weighted Cascade, and Linear Threshold diffusion models. The system includes the implementation of algorithms based on Monte Carlo simulations (Greedy, CELF, CELF++) as well as on scoring measures (centrality, PageRank, RNR). The thesis also includes experiments based on comparative analysis of real and synthetic datasets, which examine the effectiveness, computational complexity, propagation of influence, parameter sensitivity, and the effect of network structure on the speed and spread of infection. The experiments demonstrate that simulation-based algorithms are more stable and reliable than heuristics, but are computationally intensive. Their use is therefore suitable for smaller and clustered networks; for large and densely connected networks, the use of heuristics such as PageRank is much more appropriate in terms of computational and time efficiency.