LG-Index: Learning-based graph indexing for subgraph queries

Subgraph query processing is a classic problem in graph data management and has a variety of real-world applications. Existing work mainly takes the indexing-filtering-verification (IFV) approach and the index-free approach. Although a lot of effort has been made to solve subgraph queries, almost no one considers the role of historical query workloads in subgraph querying algorithms. In recent years, machine learning (ML) has shown strong performance in many research areas, but existing subgraph querying algorithms have not yet applied ML methods. In this paper, we propose a historical query-driven and data-driven learning-based graph indexing (LG-Index) framework. In LG-Index, we build indexes based on historical query workloads and the database, and speed up query processing with two learning components. We conduct extensive experiments on four real-world datasets and one synthetic dataset. The experimental results demonstrate that our approach is not only more efficient than the state-of-the-art IFV approaches by up to 67% but also can speed up recent index-free subgraph query approaches by up to 81%. Our code is available at https://github.com/BinYang121/LG-Index.