Retrieving Large-scale Product Knowledge by Collaborative Computing of CPUs and GPUs
Hao Song, Chuangxin Fang, Zhengguo Huang, Yuming Lin · 2021
The knowledge graph is the essential infrastructure of plenty of intelligent Web applications. There are various types of knowledge graphs designed and deployed to make the applications smarter during the past decade. However, the increasing amount of product data brings new challenges to the query and retrieval of products. In this work, we present a product knowledge representation method that converts text into IDs to reduce the data transmission time. In order to decrease the retrieval time, we transform the join operation in the relational database into a matrix operation. We propose a pipeline query optimization strategy under GPUs to speed up the execution of the query. To evaluate the performance, we compare our method with the state-of-the-art RDF engine RDF-3X and gStore on the large-scale product datasets. The experimental results show that our approach can significantly improve the efficiency of the SPARQL query.