【摘要】 针对隐患的多关联关系和隐患事实的时序变化特性，采用时序知识图谱的增量学习方式，针对物流企业的隐患问题，完成特定领域的隐患预测。一方面通过不同时间窗下的三元组关系，将物流隐患相关的多类别数据连接起来形成一个结构化的关系网络，充分利用现有数据事实；另一方面考虑到事实的时序变化性，以实体和时间的嵌入表达实现时间信息的融合。同时，在完成基础模型训练后，采用增量学习结合正负样本训练的方式，降低资源损耗。实验结果表明，上述算法能在保证预测准确性的基础上，大大提高模型的预测效率。更多还原

【Abstract】 In view of the multiple correlation relationships of hidden dangers and the time sequence change characteristics of hidden dangers facts, the incremental learning method of time sequence knowledge graph is adopted to complete the hidden dangers prediction in specific fields for hidden dangers of logistics enterprises. On the one hand, through the triplet relationship under different time Windows, the multi-class data related to logistics hazards are connected to form a structured relationship network, making full use of the existing data facts. On the other hand, considering the temporal variation of facts, the time information is fused by the embedded expression of entity and time. At the same time, after completing the basic model training, the method of incremental learning combined with positive and negative sample training is adopted to reduce resource loss. Experimental results show that the proposed algorithm can greatly improve the prediction efficiency of the model on the basis of guaranteeing the accuracy of prediction.更多还原