[Paper Review] DANI: A Fast Diffusion Aware Network Inference Algorithm

Maryam Ramezani, Hamid R. Rabiee+2 more|arXiv (Cornell University)|Jun 3, 2017

Complex Network Analysis Techniques41 references3 citations

TL;DR

DANI is a novel network inference algorithm that reconstructs social network topologies from diffusion cascade data while preserving community structure, using a diffusion-aware, community-preserving optimization framework. It achieves higher accuracy and significantly faster runtime than state-of-the-art methods like NETINF, DNE, and MultiTree across real and synthetic networks.

ABSTRACT

The fast growth of social networks and their privacy requirements in recent years, has lead to increasing difficulty in obtaining complete topology of these networks. However, diffusion information over these networks is available and many algorithms have been proposed to infer the underlying networks by using this information. The previously proposed algorithms only focus on inferring more links and do not pay attention to the important characteristics of the underlying social networks In this paper, we propose a novel algorithm, called DANI, to infer the underlying network structure while preserving its properties by using the diffusion information. Moreover, the running time of the proposed method is considerably lower than the previous methods. We applied the proposed method to both real and synthetic networks. The experimental results showed that DANI has higher accuracy and lower run time compared to well-known network inference methods.

Motivation & Objective

• To address the limitation of existing network inference methods that fail to preserve key structural properties like community structure in inferred networks.
• To develop a scalable and efficient inference algorithm that leverages diffusion cascade data (infection times) to reconstruct accurate network topologies.
• To ensure that inferred networks maintain realistic community-level properties such as degree distribution, clustering coefficient, and modularity for downstream applications.
• To reduce computational cost while improving inference accuracy compared to existing methods like NETINF, DNE, and MultiTree.
• To enable reliable community detection on inferred networks by preserving structural fidelity.

Proposed method

• DANI formulates network inference as an optimization problem that incorporates community structure constraints into the inference process.
• It uses a random process model to simulate diffusion dynamics and aligns inferred network links with observed infection time cascades.
• The algorithm applies a community-preserving regularization term that encourages dense internal connections within inferred communities and sparse connections across communities.
• It employs a greedy optimization strategy to iteratively refine the inferred network structure based on cascade observations and community fidelity.
• DANI integrates community detection as a validation step, comparing inferred and ground-truth communities using metrics like conductance and modularity.
• The method is designed to be cascade-length independent, ensuring consistent performance regardless of cascade duration.

Experimental results

Research questions

• RQ1 Can network inference algorithms preserve the community structure of underlying social networks when only diffusion cascade data is available?
• RQ2 How does the inclusion of community structure constraints affect the accuracy and efficiency of network inference?
• RQ3 To what extent does DANI outperform existing methods like NETINF, DNE, and MultiTree in reconstructing network topology and preserving structural properties?
• RQ4 Does the performance of DANI remain stable across different cascade lengths and network types?
• RQ5 Can community detection be reliably applied to inferred networks if structural fidelity is maintained?

Key findings

• DANI achieved a 15.84% improvement in F1-score over DNE, 94.83% over MultiTree, and 14.48% over NETINF on the News of the World dataset.
• On the NetScience dataset, DANI reduced the difference in community count between inferred and original networks by 12.8% compared to the next best method.
• DANI achieved a 29.08% average improvement in PWF (Precision of Weighted F1) over DNE on the LinkedIn dataset, with 67.72% improvement over NETINF.
• The inferred network by DANI had a density (0.0232) much closer to the original network (0.0403) than DNE (0.0433), indicating better structural fidelity.
• Conductance values for DANI were consistently close to the original network, with a variance of 0.00006, showing stable community quality.
• DANI outperformed all baselines in preserving clustering coefficient and degree distribution, with a 35.9% lower deviation in clustering coefficient on NetScience compared to DNE.