A sequence to formula tree model for solving electrical text problems

A key challenge in solving electrical text problems is to automatically adapt to understanding semantically diverse text problems. However, existing solution methods either focus on semantic or structural analysis, and they cannot parse the problem text into a more interpretable form of machine solver. A general electrical problem solver should be able to understand complex and various problem texts and apply relevant electrical theorems at the same time. In this paper, we propose a sequence to formula tree model by using a deep learning network that can parse the problem text into a formula tree with numerical values and electrical theorems. A pair of encoder and decoder of Gate Recurrent Unit is proposed to establish a formula tree graph structure for the problem text and theorem library, and the implicit electrical theorems can be extracted by using a Graph Convolutional Neural Network. Based on the formula tree and implicit electrical theorems, an electrical problem solver is proposed to produce the solutions to electrical problems. The experimental results demonstrate the effectiveness of the proposed model in solving electrical text problems. In addition, a dataset containing 3027 electrical text problems is collected and subjected to format normalization and information annotation operations.