基于Spark的OWL语义规则并行化推理算法

随着语义网的快速发展，语义数据也高速增长，传统单机推理系统无法满足推理需求，而已有的并行推理算法在推理完备性和稳定性上存在明显不足。提出的基于Spark的并行推理算法(PROS)从以下3点进行了优化：(1)通过分析OWL Horst规则依赖关系，结合数据的分类结果将规则分四类。(2)四类规则分别设计了区域最优的规则执行顺序，进一步提高了并行推理的执行效率。(3)将Sameas规则考虑到迭代中，显著提高了算法的推理能力。实验结果表明，相比已有并行推理算法，PROS并行推理算法在保证推理完备性和稳定性上表现更加出色，推理效率亦有小幅提高；同时PROS相比单机推理算法大大缩短了推理时间，处理大规模数据展现出优良的并行扩展性。

PROS: Parallel Reasoning for OWL semantic rules based on Spark

With the rapid development of semantic web, the amount of semantic data increases rapidly, which results in poor efficiency of single-node reasoning systems. On the other side, although efficient in reasoning time, the existing parallel reasoning algorithms generally show unsatisfactory reasoning integrity and stability. In this paper, we proposed an improved parallel reasoning schema, named as PROS, which includes three major optimizations as follows. Firstly, by analyzing the mutual dependencies, the OWL Horst rules are divided into four classes. Secondly, locally optimal strategies are designed for the four classes of rules, which furtherly improves the reasoning efficiency. Thirdly, we take the Sameas rules into iteration so as to boost reasoning ability of PROS. The experimental results indicate that the proposed PROS outperforms in reasoning integrity and stability compared to existing parallel reasoning algorithms. Meanwhile the PROS greatly reduces the reasoning time compared to single-node implementation and shows good expansibility.