【摘要】 电力设备故障缺陷知识图谱能够有效提升设备运维的智能化、自动化水平，而实体及关系的抽取对图谱的构建至关重要。然而故障缺陷文本中的实体关系三元组往往互相重叠或嵌套，使得传统方法难以处理，并伴随着误差传递、冗余实体推断等问题。针对这些问题，文中提出了一种面向电力设备故障缺陷领域的实体及关系联合抽取方法。该方法将三元组抽取任务建模为不同关系类型下头实体映射到尾实体的过程，通过首先抽取出头实体，再为已识别头实体针对每一种关系分别标记其对应的尾实体，从而有效缓解了三元组重叠嵌套及冗余推断等问题。实验表明，所提出的方法相较于基线模型在三元组出现不同程度重叠或嵌套时表现地更加鲁棒，其F₁值提升了8.57%～25.19%，验证了所提模型的有效性与可行性。更多还原

【Abstract】 Fault or defect knowledge graphs of power equipment could effectively improve the intelligent and automatic level of operation and maintenance of the equipment and, horever, the extractions of text and relations is crucial to the construction of the graph. However, in fault or defect text,the entity relationship triples often overlap or embedded each other,resulting in not only difficult treatment by traditional method,but also accompanying such issues as error transmission and redundant entity inference. As for these issues,in this paper a joint extraction method based on relation-oriented with Bi-LSTM is proposed.In this method,the triplet extraction task is modeled as the process of mapping the head entity to the tail entity under different relationship types. First, the head entity is extracted and then the corresponding tail of each type of relationship is marked for the identified entity,thus effectively mitigating the issues related to overlapping and nesting triples as well as redundant entity-relationship inference.The experiments show that the proposed method is more robust than the baseline model when triples overlap or nest occurring at different degrees,the F₁ value is improved by 8.57% to 25.19%,which verifies both effectiveness and feasibility of the proposed model.更多还原