Abstract:In this paper, we study several high-order network-influence-propagation frameworks and their connection to the classical network diffusion frameworks such as the triggering model and the general threshold model. In one framework, we use hyperedges to represent many-to-one influence -- the collective influence of a group of nodes on another node -- and define the hypergraph triggering model as a natural extension to the classical triggering model. In another framework, we use monotone Boolean functions to capture the diverse logic underlying many-to-one influence behaviors, and extend the triggering model to the Boolean-function triggering model. We prove that the Boolean-function triggering model, even with refined details of influence logic, is equivalent to the hypergraph triggering model, and both are equivalent to the general threshold model. Moreover, the general threshold model is optimal in the number of parameters, among all models with the same expressive power. We further extend these three equivalent models by introducing correlations among influence propagations on different nodes. Surprisingly, we discover that while the correlated hypergraph-based model is still equivalent to the correlated Boolean-function-based model, the correlated general threshold model is more restrictive than the two high-order models. Our study sheds light on high-order network-influence propagations by providing new insight into the group influence behaviors in existing models, as well as diverse modeling tools for understanding influence propagations in networks.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is, in network influence spreading, the relationships among different high - order network influence frameworks and their connections with classical network diffusion frameworks (such as the triggering model and the general threshold model). Specifically, the paper studies several high - order network influence spreading frameworks, including the hypergraph triggering model and the Boolean - function - based triggering model, and explores the equivalences between these models and the classical triggering model and the general threshold model.
### Main Contributions
1. **Model Equivalence**
- The paper proves that the Boolean - function - triggering model, the hypergraph - triggering model, and the general threshold model are equivalent under the condition of node independence. This means that these three models have the same expressive power when describing network influence spreading.
- Mathematically, the classical triggering model is strictly inferior to the general threshold model because its influence spreading is sub - modular, while the general threshold model can capture the complementarity in influence.
2. **Related Extensions**
- The paper further studies the stochastic extension versions of these models, that is, allowing correlations in influence spreading among different nodes. The results show that although the stochastic extension version of the Boolean - function - triggering model (SBFD model) and the stochastic extension version of the hypergraph - triggering model (SHD model) are equivalent, the related general threshold model is a more strict subclass.
- This finding indicates that when the stochastic behaviors among nodes are not independent, the behaviors of the general threshold model and its related extension models are no longer equivalent to other related diffusion models.
### Technical Details
- **General Threshold Model**: Each node \(v\) has a monotonic threshold function \(f_v: 2^{N^{-}(v)} \to [0, 1]\), and a random threshold \(\theta_v\in[0, 1]\). Node \(v\) is activated when the \(f_v\)-value of its active neighbors exceeds its threshold \(\theta_v\).
- **Triggering Model**: Each node \(v\) has a trigger set distribution \(T_v\), indicating which neighbors can directly influence \(v\). Node \(v\) is activated when there are any active nodes in its trigger set.
- **Hypergraph Triggering Model**: Use hyper - edges to represent the collective influence of multiple nodes on a single node. Each node \(v\) has a hyper - edge distribution, indicating which sets can activate \(v\).
- **Boolean - function - Triggering Model**: Use monotonic Boolean functions to represent the logical influence among nodes. Each node \(v\) has a Boolean activation function \(g_v:\{0, 1\}^{V\setminus\{v\}} \to \{0, 1\}\), indicating how \(v\) is influenced by other nodes.
### Related Work
- **Kempe et al.**: They first summarized and proposed several influence spreading models, including the independent cascade model and the linear threshold model, and extended these models to the general threshold model and the general cascade model, and proved their equivalences.
- **Scalable Influence Maximization**: It studies how to design fast algorithms to deal with the influence maximization problem in large - scale graphs, such as the Reverse Influence Sampling (RIS) method.
- **Other Directions**: Including competitive and complementary influence spreading, seed set minimization, profit maximization, online influence maximization, etc.
### Conclusion
The paper provides new insights into high - order network influence spreading by studying the equivalences among the Boolean - function - triggering model, the hypergraph - triggering model, and the general threshold model. These models can not only model group influence behaviors more directly, but also highlight the subtle differences in the expressive power of these general network influence models.