todo314 / dynamic-influence-analysis

Dynamic Influence Analysis

This is a real-time fully-dynamic index data structure designed for influence analysis on evolving networks.

Usage

To use our index algorithm, please include "dim.hpp".

Index construction

Our index can be build as follows:

#include "dim.hpp"
DIM dim;
dim.init();
for all vertex v
	dim.insert(v); // Insert vertex v
for all edge (u, v)
	dim.insert(u, v); // Insert edge (u,v)
dim.set_beta(beta); // Set β (See paper)

Dynamic updates

• insert(int v): Insert vertex v
• erase(int v): Delete vertex v (and its incident edges)
• insert(int u, int v, double p): Insert edge (u,v) with probability p
• erase(int u, int v): Delete edge (u,v)
• change(int u, int v, double p): Change probability of edge (u,v) to p

Influence queries

• infmax(int k): Extract an influential set of k vertices
• infest_naive(vector<int> &S): Estimate the influence of S (slow implementation)
• infest(vector<int> &S): Estimate the influence of S (fast implementation)
• infest(int v): Estimate the influence of v

Example

#include "dim.hpp"
DIM dim;
dim.init();

// index construction
dim.insert(0);
dim.insert(1);
dim.insert(0, 1, 0.4);
dim.set_beta(32);

// dynamic updates
dim.insert(2);
dim.insert(3);

dim.insert(1, 2, 0.5);
dim.insert(2, 0, 0.6);
dim.insert(2, 3, 0.7);

// influence queries
dim.infest(0); 
dim.infmax(1);

References

• Naoto Ohsaka, Takuya Akiba, Yuichi Yoshida, and Ken-ichi Kawarabayashi. Dynamic Influence Analysis in Evolving Networks. Proceedings of the VLDB Endowment, 9(12), pages 1077–-1088, 2016.