Measuring the effects of repeated and diversified influence mechanism for information adoption on Twitter

Social Network Analysis and Mining (2022) 12:16 https://doi.org/10.1007/s13278-021-00844-x REVIEW PAPER Measuring the effects of repeated and diversified influence mechanism for information adoption on Twitter Jaqueline Faria de Oliveira1,3 · Humberto Torres Marques‑Neto1 · Márton Karsai2,3 Received: 28 February 2021 / Revised: 14 October 2021 / Accepted: 12 November 2021 © The Author(s), under exclusive licence to Springer-Verlag GmbH Austria, part of Springer Nature 2021 Abstract People can adopt information disseminated in online social networks whenever they receive it frequently from friends or others. Studying this social influence dynamic is crucial to understanding social interactions and users’ behavior regarding online information spread. Quantifying social influence is challenging in online social systems where the interactions and communication content can be closely followed. Here, we study the effects of repeated and diversified influence mechanisms exploring the concepts of Information susceptibility and Adoption thresholds of Twitter users. We consider hashtag and retweet adoptions on different aggregation levels: items, users, and topic groups and study the adoption characterized by diversified and repeated influence stimuli. We address this challenge by tracking the timeline order of potential influence and adopting hashtags and retweets in a specific dataset collected from Twitter, which contains the posts’ dynamics of thousands of seed users and their entire followee networks. We show that users adopt retweets easier than hashtags, and we find both metrics to be heterogeneously distributed, correlated, and dependent on the topics and aggregation level of social influence. We find that new influencing neighbors can effectively trigger adoptions, particularly for topics where a new adopter friend triggers ~ 50% of adoptions. Our results may inform better models of adoption processes leading to a deeper empirical understanding of simple and complex contagion in online social networks. Keywords Susceptibility · Threshold · Social contagion · Adoption 1 Introduction Think about how you decide to adopt some behavior or information. Would you adopt information easier if you are exposed to it many times or if many different friends would tell you the same? This is a central question in social influence studies where different adoption decision mecha‑ nisms have been identified behind these two scenarios. In * Jaqueline Faria de Oliveira Humberto Torres Marques‑Neto Márton Karsai 1 Department of Computer Science, PUC Minas, Belo Horizonte, Brazil 2 Department of Network and Data Science, Central European University, Vienna, Austria 3 ENS de Lyon, Inria, CNRS, UCB Lyon 1, LIP UMR 5668, IXXI, Univ Lyon, 69342 Lyon, France fact, decisions cannot be associated clearly with one of these mechanisms but probably have combined effects on one’s decisions. In this paper, our goal is to observe differ‑ ences between these scenarios by measuring the repeated, and diversified influence mechanism using information sus‑ ceptibility and adoption thresholds metrics of hashtags and retweets on different levels, using a large Twitter dataset collected for this purpose. Influence is an effect emerging between socially engaged people where one person may change behavior just by observing what a peer or a group is doing, to become more similar to them, to increase acceptance, and decrease social pressure. Social influence may be unintentional and can also be used to intentionally induce behavioral change as com‑ monly exploited in business and propaganda. This is espe‑ cially true in the digital data revolution era, where online social platforms and news sites have been commonly used to bias public opinion or to sell products and services, to mention a few examples. The adoption of items like information, products, or any behavioral pattern is partially driven by social influence and 13 Vol.:(0123456789) 16 Page 2 of 15 may be interpreted as a spreading process between people connected in a social structure. While some of items can be easier adopted, like information, memes, or hashtags (which are in the main focus of this paper), the adoption of other types of patterns, like buying a product, join a riot, or change opinion, is more cumbersome due to their higher cost (being financial, social, or intellectual) for the individual. The spreading of adoptions has been modeled in various ways over the last decades. Early models by Bass (1969) and Rogers (2010) assume that it spreads similarly to biological epidemics, where the number of incoming stimuli from any peer may increase the probability of adoption. In the social contagion terminology, this is called simple spreading. On the other hand, an alternative modeling paradigm has been proposed based on the arguments of Schelling (1969), Gran‑ ovetter and Soong (1983) and Watts (2002), who identified adoption as a threshold driven process where not the number of exposure but the number or fraction of earlier adopted friends in one’s egocentric network determines adoption. They argue that each person has a cognitive threshold of influence, expressed as a fraction of adopted peers, which has to be succeeded for the adoption to take place. Adop‑ tion spreading driven by threshold mechanisms are called complex spreading processes. To characterize the effects of different influence mecha‑ nisms one could devise different measures. Susceptibility (Hoang and Lim 2012) is defined as the fraction of the num‑ ber of adoption and exposure to a set of items propagated multiple times from the peers of an ego. This metric captures well the effects of simple contagion as it measures how the probability of adoption changes via multiple stimuli com‑ ing even from a single neighbor. Adoption threshold (Watts 2002), on the other hand, measures the fraction of already adopted and the total number of friends at the time of a single adoption of an ego. This metric captures directly indi‑ vidual thresholds assumed to drive adoption spreading as suggested by complex contagion. Due to recently emerging popular online social plat‑ forms and systematically collected purchase data sets, it is now relatively easy to follow adoption behavior in some cases. The micro-blogging service of Twitter provides pre‑ cise records of users’ action in adopting or sharing infor‑ mation, while it also records the potential social structure facilitating social influence between them. However, even the differences between the two discussed adoption mecha‑ nisms are easy to understand, and we have data provid‑ ing precise observations of adoptions, it is surprisingly difficult to distinguish the effects of simple and complex contagion in local and global observations. On one hand, the strength of social influence between people may vary considerably (Cox et al. 2016; Hurd and Gleeson 2013; Unicomb et al. 2018), and an ego may have different inter‑ est to ad (...truncated)