一种基于图文理解的电路题目自动解答方法

Automatic Solution Method for Circuit Problems based on Graphic-Textual Understanding

自动解答是人工智能领域的一个研究热点。由于题目文本中包含自然语言描述的复杂情景,电路图形中亦包含众多的电路结构信息,因此图文理解成为自动解答中的一个关键挑战。针对该挑战,提出了一种基于图文理解的电路题目自动解答方法。该方法将图文理解的过程抽象为一个关系抽取的过程,并提出了一种句法语义模型用于抽取题目文本中的关系,进而提出一种电路网孔搜索法与深度卷积神经网络相结合的方法用于抽取电路图形中的关系,并求解抽取的关系,以实现电路题目的自动解答。通过对电路题目的解答实验,句法语义模型抽取了93.5%的文本关系,网孔搜索法对电路图形中的VCR、KCL和KVL关系抽取正确率分别达到88.2%、90.9%和80.7%,最终实现了75.56%的解答正确率。

Automatic solution of circuit problems is a research hotspot in the field of artificial intelligence.Due to the fact that the problem text contains complex scenarios described in natural language,and the circuit graph also contains a lot of circuit structure information,graph-text understanding,therefore,becomes a key challengein automatic solution.Aiming at this challenge,an automatic solution mrthod for circuit problems based on graphtext understanding is proposed.This methodabstracts the process of graph-text comprehension into a process of relations extraction.A syntax-semantics model is proposed to extract relations in the problems text,and then a circuit mesh search method in combination with deep convolutional neural network is further proposed,so as to extract the relations in the circuit graph and solve the relations of extraction,thus realizing the automatic solution of circuit problems.Experimental results indicate that the syntax-semantics model could achieve an accuracy of 93.5% in relations extraction from text,and the accuracies of extracting VCR,KCL and KVL relations coule reach88.2%,90.9% and 80.7% respectively from circuit graphby mesh search method.Finally,the correct solution rate of 75.56% circuitproblems is achieved by the proposed method.