基于知识表示学习的实时语义数据流推理

传统的语义数据流推理使用前向或后向链式推理产生确定性的答案,但是在复杂的传递规则推理中效率不高,无法满足实时数据流处理场景对答案的及时性要求。因此,提出一种基于联合嵌入模型的知识表示方法,并应用于语义数据流处理中。将规则与事实三元组联合嵌入并利用深度学习模型进行训练,在推理阶段,根据查询中涉及的规则建立推理模板,利用深度学习模型对推理模板产生的三元组进行预测和分类,将结果作为查询和推理答案输出。实验表明,对于复杂规则推理,基于知识表示学习的实时语义数据流推理能够在保障较好推理准确性和命中率的前提下有效地降低延迟。

REAL-TIME SEMANTIC DATA STREAM REASONING BASED ON KNOWLEDGE REPRESENTATION LEARNING

Traditional semantic data stream reasoning methods use forward or backward chained reasoning to generate deterministic answers,but it is inefficient in complex reasoning with transitive rules,and cannot satisfy the timeliness requirements in real-time data stream processing scenarios.Therefore,this paper proposes a knowledge representation method based on the joint embedding model to apply in semantic data stream processing.It embedded rules and fact triples jointly and used deep learning models for training.In the inference phase,target inference templates were produced based on the rules involved in the query.The deep learning model was used to predict and classify the triples generated by the inference template,and derived the results as reasoning answers for the RDF stream processing engine.The experiments show that for complex rules,real-time reasoning based on knowledge representation method can effectively reduce latency while guaranteeing acceptable inference accuracy and hit rate.