A Sample Reuse Strategy for Dynamic Influence Maximization Problem (2311.15345v1)

Abstract: Dynamic influence maximization problem (DIMP) aims to maintain a group of influential users within an evolving social network, so that the influence scope can be maximized at any given moment. A primary category of DIMP algorithms focuses on the renewal of reverse reachable (RR) sets, which is designed for static social network scenarios, to accelerate the estimation of influence spread. And the generation time of RR sets plays a crucial role in algorithm efficiency. However, their update approaches require sequential updates for each edge change, leading to considerable computational cost. In this paper, we propose a strategy for batch updating the changes in network edge weights to efficiently maintain RR sets. By calculating the probability that previous RR sets can be regenerated at the current moment, we retain those with a high probability. This method can effectively avoid the computational cost associated with updating and sampling these RR sets. Besides, we propose an resampling strategy that generates high-probability RR sets to make the final distribution of RR sets approximate to the sampling probability distribution under the current social network. The experimental results indicate that our strategy is both scalable and efficient. On the one hand, compared to the previous update strategies, the running time of our strategy is insensitive to the number of changes in network weight; on the other hand, for various RR set-based algorithms, our strategy can reduce the running time while maintaining the solution quality that is essentially consistent with the static algorithms.