Grunnleggende konsepter
Proposing MOEIM for Many-Objective Evolutionary Influence Maximization, outperforming competitors in various settings.
Sammendrag
Introduction:
Influence Maximization (IM) problem seeks to identify nodes for maximum information spread.
NP-hard problem studied with various objectives like seed set size, fairness, budget, and time.
Background:
IM problem defined as combinatorial optimization over a graph.
Propagation models like IC, WC, LT used for influence spread simulations.
Related Work:
Works focus on influence, seed set size, fairness, communities, budget, and time objectives.
Methods:
MOEIM introduced with smart initialization, evolutionary optimization, and graph-aware operators.
Experimental Setup:
Two experimental settings comparing MOEIM with heuristics, MOEA, and DeepIM on various datasets.
Results:
MOEIM outperforms competitors in most cases, especially in multi-objective settings.
Correlation Analysis:
Hypervolume results show interesting patterns in optimizing different combinations of objectives.
Statistikk
The Influence Maximization (IM) problem is known to be NP-hard.
MOEIM outperforms competitors in most tested many-objective settings.
MOEIM significantly outperforms the competitors in terms of achieved hypervolumes.
Sitater
"MOEIM overall outperforms the competitors in most of the tested many-objective settings."
"The experiments show that MOEIM significantly outperforms the competitors in terms of achieved hypervolumes."