COEVOLVE: A Joint Point Process Model for Information Diffusion and Network Co-evolution (1507.02293v2)

Abstract: Information diffusion in online social networks is affected by the underlying network topology, but it also has the power to change it. Online users are constantly creating new links when exposed to new information sources, and in turn these links are alternating the way information spreads. However, these two highly intertwined stochastic processes, information diffusion and network evolution, have been predominantly studied separately, ignoring their co-evolutionary dynamics. We propose a temporal point process model, COEVOLVE, for such joint dynamics, allowing the intensity of one process to be modulated by that of the other. This model allows us to efficiently simulate interleaved diffusion and network events, and generate traces obeying common diffusion and network patterns observed in real-world networks. Furthermore, we also develop a convex optimization framework to learn the parameters of the model from historical diffusion and network evolution traces. We experimented with both synthetic data and data gathered from Twitter, and show that our model provides a good fit to the data as well as more accurate predictions than alternatives.

Overview of COEVOLVE: A Joint Point Process Model for Information Diffusion and Network Co-evolution

The paper introduces Coevolve, a sophisticated generative model designed to capture the intricate dynamics of co-evolving information diffusion and network evolution processes. By leveraging the framework of temporal point processes, this model offers a nuanced representation of how information flow can influence and be influenced by the topology of social networks. Such a model addresses the traditional limitation in the paper of networked systems, where the dynamic processes of information diffusion and network evolution have predominantly been examined in isolation despite their inherently interconnected nature.

Model Components

The Coevolve model hinges on two intertwined components: the information diffusion process and the network evolution process. These components are critical for capturing the bidirectional influence between network topology and information propagation.

1. Information Diffusion Process: The authors propose a multivariate Hawkes process tailored to capture the retweeting behavior in online networks. This process acknowledges the mutual excitations driven by a user's social environment. A key aspect is its "identity revealing" characteristic, which maintains the explicit acknowledgment of information sources and adjusts retweet probabilities based on perceived informativeness.
2. Network Evolution Process: Link creation is represented as an information-driven survival process. Here, the retweet events drive new connections, reflecting observed behaviors where user interaction and shared content prompt new link formations.

Simulation and Parameter Learning

The Coevolve model stands out not only in its theoretical formulation but also in its practical implementation. The paper describes an efficient algorithm for simulating co-evolutionary dynamics, accounting for the bidirectional influences between the network and the diffusion process. By exploiting network sparsity and using a scalable sampling procedure, the model achieves complexity that scales favorably with the size of the network.

Parametrically, the authors utilize a convex optimization framework for parameter estimation, ensuring the model can be trained effectively on historical data. This aspect is pivotal for aligning the model's predictions with empirical observations and for making forecasts about future dynamics.

Experimental Validation and Findings

Through rigorous experiments on both synthetic and real-world Twitter datasets, the Coevolve model demonstrated significant predictive accuracy over existing models. It was able to capture the co-evolutionary dynamics, producing realistic network patterns such as cascade structures, degree distributions, and triadic closures. Moreover, it provided more accurate predictions for link creation and information diffusion events than competitive approaches.

Implications and Future Directions

The Coevolve model contributes to the broader understanding of complex networked systems, offering insights into the interplay between social network structure and information dynamics. Practically, it could inform the design of interventions or strategies in digital marketing, information dissemination, and cyber-security by predicting where and how information is likely to spread.

Theoretically, this work opens the door to multiple avenues of future research, such as incorporating additional node features, supporting time-varying parameters, or extending the framework to support scenarios of link deletion and node churn. Each extension would further refine the model's applicability and enhance its capacity to mirror the sophisticated nature of real-world social systems.

In summary, Coevolve is a noteworthy advancement in modeling complex, intertwined social processes, providing both a robust theoretical foundation and a versatile practical tool for researchers examining large-scale social interactions.