Conformity-aware adoption maximization in competitive social networks

Abstract

Influence maximization (IM) problem is an extensively studied problem in social networks. It aims to find a small set of users in the social network to initiate the diffusion process and maximize the expected influence spread. Existing works on conformity-aware IM focus on the interaction between influence and conformity in a single-influence setting and ignore the role of conformity in a competitive and multiple-influence setting. This paper proposes a conformity-aware independent cascade (C-IC) model that considers the competition among multiple influences as well as the role of conformity in a user’s decision-making. It is proved that the adoption of an influence under the C-IC model is monotone and submodular. Meanwhile, we formulate two adoption maximization (AM) problems, O-AM and S-AM, which are both NP-hard. Because estimating the adoption through diffusion simulations is very time-consuming, we propose a reverse adoption estimation (RAE) method based on a reverse multiple influence sampling (RMIS) technology for the C-IC model and integrate it into the D-SSA-fix (Nguyenet al., 2018) framework, DSSA for short, to compute a solution with approximation guarantee. To further boost the performance, we present a fast one-hop adoption estimation (OAE) method and develop a heuristic algorithm based on OAE, called GOAE. Extensive experiments on eight real-world social networks show that the C-IC model is superior to a non-conformity diffusion model and that RAE+DSSA and GOAE are efficient and effective. In most cases, GOAE finds comparable solutions to RAE+DSSA and CELF with less time and memory overhead. GOAE is five to six orders of magnitude faster than CELF and RAE+DSSA is up to three orders of magnitude faster than CELF on NetHEPT. GOAE runs up to four to five orders of magnitude faster than RAE+DSSA with at most two orders of magnitude less memory usage. GOAE is more scalable than RAE+DSSA in terms of the number of seeds and the size of the social network.