基于转移的航空安全事件知识图谱表示学习

Knowledge graph representation learning of aviation safety incidents based on transfer

通过构造航空安全事件知识图谱并对其进行推理预测,可以有效预防航空安全事件的发生。目前,对于知识图谱的表示学习大多采用转移模型TransE,虽然其具有简单、高效的优势,但是在处理复杂关系时存在局限性。航空安全事件知识图谱不同于其他领域知识图谱,其中每个事件相互独立且又联系紧密,存在大量复杂关系,TransE模型不能很好地对其进行表示学习。为此,通过对航空安全事件语料库进行抽取来构建ASIKG数据集,利用公开数据集和ASIKG数据集对TransE的改进模型进行训练,实验结果表明,TransR模型在公共数据集上链接预测效果较好,而TransH模型在ASIKG数据集上取得了较好的链接预测效果。

By constructing the knowledge graph of aviation safety incidents and inferring it,it can effectively prevent the occurrence of aviation safety incidents.At present,transfer model TransE is mostly used for representation learning of knowledge graph.Although it has advantages of simplicity and efficiency,it has limitations when dealing with complex relationships.Aviation safety incident knowledge graph is different from other domain knowledge graph,in which each event is independent and closely related,there are a lot of complex relations,and the TransE model can not well represent and learn it.To this end,the ASIKG data set was constructed by extracting the aviation safety incident corpus,and the improved TransE model was trained by using the open data set and ASIKG data set.The experimental results show that the TransR model has a good link prediction effect on the public data set.The TransH model has a good link prediction effect on the ASIKG dataset.