Generating Graphs With Specified Properties And Their Use For Constructing Scene Graphs From Images

Abstract

Graph generation, the process of creating meaningful graphs, plays a vital role in various domains, including social network analysis, bioinformatics, recommendation systems, and network modeling. This article provides three graph generation models and also proposes the idea of constructing a scene graph using graph generation models. The where different models graph generation has been used for purposes such as social network analysis for community discovery, bioinformatics for protein interaction networks, recommendation systems for personalized recommendations, and network modeling for simulating real-world scenarios. In such models, the hidden state matrix of generated objects was used as a feature matrix. This article sets the goal of building a model with the ability to generate various types of graphs, without being tied to a specific area of application, that is, a matrix describing the structural characteristics of graphs will be used as a feature matrix. This paper develops three methods for generating graph structures with given properties using generative neural networks. The developed methods are tested on the set of Hamiltonian graphs. A comparative analysis of the quality of the generated graph structures is performed. A method of scene graph construction using the developed methods is proposed.