End-to-End Community Search based on Graph Transformer: A Demonstration

Zhiyuan Yu, Yuxiang Wang, Xiaoliang Xu, Xiaoxuan Gou

Abstract: Given a graph G and a query vertex q, Community Search (CS) aims to find a cohesive subgraph from G that contains q. The advent of deep learning has led to the emergence of various learning-based methods for CS, which can largely be categorized into graph neural network (GNN)-based and transformer-based methods. The GNN-based methods gather information from neighbor vertices accurately and efficiently, but is limited by the link distance — the finite layers of GNN cannot aggregate information from vertices at distal ends of long-range. In this paper, we combine the GNNs with Transformer architecture to propose an end-to-end CS method—Query Driven Graph Transformer (QDGT), which enhances search capabilities and overcomes the deficiency in long-range reasoning and the inability to ensure structural connectivity of GNNs. Moreover, we developed a CS prototype to provide intuitive exhibition of the resultant communities, analytical outputs, and model details, to help users better understanding our model’s performance in runtime.

External IDs:dblp:conf/cbd/YuWXG23