Abstract
We postulate that the nature of information items plays a vital role in the observed spread of these items in a social network. We capture this intuition by proposing a model that assigns to every information item two parameters: endogeneity and exogeneity. The endogeneity of the item quantifies its tendency to spread primarily through the connections between nodes; the exogeneity quantifies its tendency to be acquired by the nodes, independently of the underlying network. We also extend this item-based model to take into account the openness of each node to new information. We quantify openness by introducing the receptivity of a node. Given a social network and data related to the ordering of adoption of information items by nodes, we develop a maximum-likelihood framework for estimating endogeneity, exogeneity and receptivity parameters. We apply our methodology to synthetic and real data and demonstrate its efficacy as a data-analytic tool.
Introduction
Traditional social network analysis often prioritizes network topology—who is connected to whom—and node-level characteristics like homophily. However, this approach frequently overlooks a critical factor: the inherent nature of the information being shared. Whether a piece of information is a global news event or a niche personal recommendation fundamentally alters its propagation pattern.
This paper shifts the focus to the "what" of social sharing. By characterizing information items through two primary lenses—endogeneity (internal network spread) and exogeneity (external source influence)—and accounting for individual user receptivity, we can better understand why certain memes achieve viral status while others require continuous external pressure to persist.
Modeling Information Spread: The E2 and E2R Frameworks
The authors propose two probabilistic models to capture the dynamics of information adoption:
- The E2 Model: This model assigns two parameters to each information item $i$. Exogeneity ($x_i$) represents the probability a node adopts an item due to external factors (e.g., mass media), while Endogeneity ($e_i$) represents the probability of adoption via a single active neighbor. The probability of a node $u$ becoming active for item $i$ is defined as $1 - (1 - x_i)(1 - e_i)^{\Gamma(u,i)}$, where $\Gamma(u,i)$ is the count of $u$'s active neighbors.
- The E2R Model: This extension introduces Receptivity ($r_u$), a user-specific parameter capturing a node's general openness to new information. In this model, item parameters are scaled by $r_u$, allowing the model to distinguish between an item's lack of virality and a user's inherent resistance to information.
Parameter estimation is achieved through a Maximum Likelihood Estimation (MLE) framework. While E2 parameters can be calculated in parallel, the E2R model utilizes a coordinate descent procedure. Optimization is made computationally efficient—reducing per-iteration complexity from $O(n)$ to $O(1)$ for sparse data—enabling the analysis of large-scale datasets in minutes.
Experimental Validation
The methodology was first tested on synthetic ScaleFree graphs (1,000 nodes, 1,000 items) generated using the Barabási-Albert model. The "Item-Fit" algorithm demonstrated high accuracy in recovering ground-truth parameters, even in the sparse data environments typical of real-world networks.
As shown in the heatmaps above, the error rates for both exogeneity and endogeneity remain consistently low across a wide range of input values, validating the robustness of the MLE approach.
Real-World Analysis: The MemeTracker Dataset
The models were applied to the MemeTracker dataset, encompassing over 56,000 unique quotes across nearly 50,000 blog nodes. The analysis revealed distinct patterns in how information propagates:
- Exogeneity vs. Endogeneity: Real-world exogeneity values were found to be generally small (peaking at 0.032), whereas endogeneity spanned the full range from 0 to 1.
- Hybrid Spread: High-profile quotes like "Joe the plumber" exhibited high scores in both dimensions, indicating they were driven by both intense media coverage and peer-to-peer sharing.
- The Impact of Receptivity: The E2R model refined these findings by identifying that some items appeared less "viral" simply because they reached low-receptivity nodes. For example, political slogans like "Yes we can" showed high exogeneity and high user receptivity during the 2008 election, rather than pure peer-to-peer contagion.
The distribution of endogeneity (Figure 4) highlights the diversity of information "natures," with a significant number of items showing high potential for network-driven spread.
Conclusion
This research demonstrates that the spread of information is a product of both the network's "nurture" and the information's "nature." While current models assume static parameters and uniform link influence, they provide a strong foundation for future work. Integrating Natural Language Processing to correlate textual content with these probabilistic parameters could further unlock the secrets of information dynamics in social networks.
Summary updated