Athens University Economics and Business (AUEB)
Modeling the online social transmission through the theory of complex adaptive systems and neural network models based on the brain dynamics
Abstract
dc:descriptionThe modeling of online social dynamical processes is a challenging undertaking due to the following reasons: First, the interacting individuals are heterogeneous and constitute complex systems themselves. Second, the interactions among individuals are subject to noise and stochasticity, while the network of interactions is highly complex and mainly intractable. Third, social systems exhibit a non-equilibrium behavior, as their dynamics is affected by external perturbations whose distribution, correlation with the system, and non-stationarity are difficult to capture and incorporate into a social dynamical model. The successful modeling of online social transmission processes calls for an approach capable of addressing the aforementioned challenges. To this end, I build and apply an implementation framework based on the theory of complex adaptive systems. Such a methodology captures the dynamic nature of interactions as well as the macroscopic properties of the individuals' activity in the presence of external influences.A complex adaptive system of immense complexity is that of the human brain. Social networks are even more complex as they consist of interconnected brains. As a result modeling dynamical processes occurring on the social web is a formidable task. To address the challenges of such an endeavor, I adopt a neuroscience perspective and consider the dynamics of online social networks analogous to the dynamics of networks of Integrate-and-Fire neurons. Through this isomorphism I introduce a novel online social contagion model driven by three sources of positive or negative influence: The individuals' self-generated bias, the interpersonal online interactions and the external environment stimulus. The proposed model explains the online activity growth and the activity propagation patterns in relation to the interaction network, the endogenous and exogenous influence and the individuals' activation mechanism. Through simulation experiments and validation against empirical data from Twitter, I demonstrate that the model accurately reproduces collective activity patterns arising from the users' response to various kinds of stimuli. Comparisons with state-of-the-art methods reveal the conceptual superiority of the proposed model which substantially exceeds the performance of existing methods in the replication of real online activity patterns. Another social transmission process that I model in a novel way in this Thesis, relates to the propagation of online information. The dynamics of this process is captured through a network dynamical system based on noisy spiking neurons. Through the combination of deterministic and stochastic components the model realistically reproduces the propagation patterns of online information, and indicates that these patterns depend on the temporal structure, strength, and signal-to-noise ratio characteristics of the influence affecting the interconnected online users. The proposed mechanism incorporates the notions of ``simple'' and ``complex'' contagion and extends the existing approaches through the inclusion of endogenous and exogenous, positive and negative, deterministic and stochastic sources of influence into a single model. The proposed neural network models are highly adaptable and suitable for the study of a large number of online and offline social dynamical processes pertaining to the spread of behavior, fashions and trends, word-of-mouth, adoption of new products, and marketing among others. Finally, I further exploit the insights gained through the modeling of online activity patterns by developing a predictive model capable of generating precise forecasts concerning the adoption of online content.
Degree
thesis:*- Grantor dc:publisher
- Athens University Economics and Business (AUEB)
- Year dc:date
- 2015
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Lymperopoulos, Ilias
- Λυμπερόπουλος, Ηλίας
Subjects
dc:subject × 10Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/43093