A question answering model for electrical equipment standards based on subject object attention

In order to reduce the cost of manual question and answer for electric equipment standards and to solve the difficulties in querying and extracting answers to complex questions caused by the large number and many versions of electric equipment standards, a retriever-reader question-and-answer model EES-QA (Electrical Equipment Standards Question Answer) for electric equipment standards is proposed. Firstly, it identifies the names of electrical equipment and discriminates the types of questions. Then, it constructs coarse and fine-grained electrical equipment standards graphs to precisely locate the answer context related to the questions. Lastly, it predicts the span of answers by the proposed question subject-object attention mechanism combined with masking operations. The proposed question-answer model is validated on the Electrical Equipment Dataset (EED) and the public dataset SQuAD. The experimental results show that the accuracy of extracting answers to standard questions on electrical equipment reaches 94%. In addition, the performance is similar to other advanced models on the public dataset.