Тип публикации: статья из журнала
Год издания: 2025
Ключевые слова: Uncertainty knowledge graph, rectangular embedding, geometric graph neural network, information transfer paradigm, affine transformation
Аннотация: In this paper, we study geometric embedding based graph neural networks and propose a Geometry-Aware Graph Neural Network with Box Embeddings (BoxGNN) for uncertainty knowledge graph representation learning. BoxGNN contains two main components, that is, information propagation and information aggregation. The message propagation paПоказать полностьюradigm in the rectangular embedding space can naturally describe the uncertainty knowledge graph in the rectangular space and fully utilize the graph structure information of the rectangular embedding representation and uncertainty knowledge graph that contains probabilistic semantics. Meanwhile, this paper proposes different message aggregators to update the rectangular embeddings and observes that the proposed mean-value aggregator obtains the experimental best performance.
Журнал: Системы управления и информационные технологии
Выпуск журнала: № 4
Номера страниц: 4-10
ISSN журнала: 17295068
Место издания: Воронеж
Издатель: Воронежский государственный технический университет