Towards establishing the effect of self-similarity on influence maximization in online social networks

Abstract

Most of the existing work done in the area of influence maximization (IM) in online social networks (OSNs) aims at the development of an algorithm for identification of seed set or the development of an information diffusion model for spread maximization. IM finds major application in the field of viral marketing, wherein an organization wants to maximize the spread of information about its product/service. Multiple OSNs might be available for initiating the diffusion process. The organization would then want to select an OSN that would lead to highest spread. However, no work exists that helps in assessing which OSN, from amongst the available set of OSNs, can be expected to achieve a higher influence spread. Seeking an answer to this problem, a method for identifying, which OSN shall achieve a higher spread, has been proposed in this paper. The proposed method explores the correlation between self-similar behaviour of user activity in an OSN and the expected influence spread for that OSN. Analogous to real-world human behaviour, which displays self-similarity, behaviour of users in an OSN can also be assumed to display self-similarity. The results achieved for the proposed work display a positive correlation between self-similarity in user activity in an OSN and the influence spread expected to be achieved in that OSN. Based on the findings, an algorithm has been proposed, which can be used to select which OSN, from amongst a set of OSNs, can be expected to achieve higher influence spread. Additionally, a two-step Hurst exponent (H)-based approach has also been proposed for IM, which makes use of the overall global H value for the OSN and local H value for each node in the OSN to ensure maximum influence spread.