GraphQueryML: Using Machine Learning to Optimize Queries in Graph Databases

查询优化，即将声明性查询语句转换为高效的查询执行计划，是数据库系统研究的中心问题之一。即使经过40年的研究，查询优化的许多子问题仍然没有得到解决。认识到越来越多的数据集是图结构的，特别是在资源描述框架(RDF)或属性图(PG)数据模型中表示的事实，该提案探索了使用机器学习来优化图形数据库中的查询这一重要的开放研究问题。(1)设计并开发了一个基于深度强化学习的通用查询优化框架。(2)将该框架应用于RDF数据库中SPARQL查询的优化。(3)研究属性图数据库中Cypher查询的优化。我们的方法具有巨大的潜力，无论是在科学界还是在工业中都能实现新的发现。特别是，广泛采用RDF数据库的数据密集型生物信息学社区将受益于跨多个RDF数据库的加速查询，从而缩短科学发现周期。

Query optimization, i.e., the translation of a declarative query statement into an efficient query execution plan, is one of the central problems of database systems research. Even after four decades of research many sub-problems of query optimization are still unsolved. Acknowledging the fact that an increasing number of data sets is graph-structured and, in particular, represented in the Resource Description Framework (RDF) or in the Property Graph (PG) data model, this proposal explores the important open research problem of using machine learning for optimizing queries in graph databases. (1) We will design anddevelop a general query optimization framework that uses machine learning with focus on deep reinforcement learning. (2) We apply our framework to the optimization of SPARQL queries in RDF databases. (3) We will study the optimization of Cypher queries in property graph databases. Our approach has the great potential to enable novel discoveries both in the scientific community as well as in industry. In particular, the data-intensive bioinformatics community with the wide adoption of RDF databases will be benefit from accelerated queries across multiple RDF databases and thus enable shorter scientific discovery cycles.