Page 68 - Read Online

Page 8 of 15 Ma et al. Intell Robot 2023;3(4):581-95 I http://dx.doi.org/10.20517/ir.2023.33

Figure 2. Structure diagram of GHNN-ACO algorithm model. GHNN-ACO: Graph-based Heterogeneous Neural Network Ant Colony
Optimization.

of the entire graph through information transmission and feature aggregation.

ACOalgorithmsareheuristicoptimizationalgorithmsthatsimulatethebehaviorstrategy ofantsin theprocess
ofsearchingforfood. Inthesealgorithms,theantsguideotherantstochoosethepathbyreleasingpheromones,
thus achieving the collaborative search of all ants. The pheromone concentration on the path will increase or
decrease according to the quality of the path, and ants tend to choose the path with a higher pheromone
concentration.

GHNN-ACO combines GHNNs and ACO algorithms to improve the classification, prediction, and feature
learning of graph data.

When it comes to solving complex problems in graph data, traditional machine learning algorithms and neural
networks often face challenges. The complexity of graph data lies in the fact that the relationships and features
betweennodesmaybeheterogeneous, meaningthatdifferenttypesofnodesandedgeshavedifferentattributes.
In order to effectively handle this heterogeneity, a GHNN has been introduced, which is a neural network
structure suitable for heterogeneous graph data, as shown in Figure 2.

A GHNN typically consists of multiple subnetworks, each responsible for handling a type of node or edge.
This structure allows subnetworks to focus on different types of information without confusing or ignoring