基于序列公式树模型的电学问题解答方法

doi:10.19678/j.issn.1000-3428.0068419

摘要/Abstract

摘要：

自适应理解和求解语义多变的问题文本是机器解答电学问题的关键挑战, 现有方法多侧重于问题文本的语义分析和结构分析, 无法将问题文本解析为类人解答的求解形式。为此, 构建一种基于序列公式树模型的电学问题解答方法。首先通过问题文本预处理标准化文本元素、关系提取, 生成预编码序列和直陈关系序列。其次使用双向门控循环编码器对预编码序列进行特征编码, 生成隐藏状态序列。再通过构建电学定理图, 使用图卷积神经网络(GCNN)编码器建立直陈关系序列和定理之间的关联关系, 将定理图中的关系结点转化为向量表示, 生成公式结点嵌入状态序列, 并提取不同变量之间的隐含电学关系。最后构建树形结构的解码器对隐藏状态序列和公式节点嵌入状态序列进行解码, 形成问题的序列-公式树结构的求解式, 实现电学问题的可读解答。构建一个包含3 027个电学问题的数据集TexPE-3K, 并对其进行标准化和信息标注。在数据集TexPE-3K上的实验结果表明, 关系提取的平均准确率达到了96.8%, 可读解答的平均准确率达到了55.57%, 验证了该方法的可行性和有效性。

关键词: 电学问题, 序列公式树模型, 关系提取, 可读解答, 图神经网络

Abstract:

The primary challenge in machine-based electrical question answering lies in effectively comprehending and addressing the semantically diverse nature of question text. Existing approaches predominantly emphasize semantic and structural analyses but struggle to generate human-like responses. An electrical question-answering method based on the sequence to formula tree model has been developed to address this issue. This method comprises several key steps. First, pre-coded sequences and directly extracted relation sequences are generated through standardized text elements and relation extraction during question text preprocessing. Second, these pre-coded sequences undergo feature encoding using a bidirectional gated recurrent encoder, resulting in the generation of hidden state sequences. Subsequently, an electrical theorem graph is constructed, and a Graph Convolutional Neural Network(GCNN) encoder is employed to establish the correlation relationships between the direct relation sequence and relevant theorems. This process transforms the relation nodes within the theorem graph into vector representations, facilitating the generation of formula node-embedded state sequences and the extraction of implicit electrical relationships among various variables. Finally, a tree-structured decoder is designed to decode the hidden state and formula node-embedded state sequences, resulting in the formulation of a solution equation within the sequence-formula tree structure. This solution aims to provide a readable resolution to the electrical problem by meticulously constructing, standardizing, and labeling the TexPE-3K dataset, comprising 3027 electrical problems. Through experiments conducted on the TexPE-3K, the method consistently achieves an average accuracy of 96.8% in relation extraction and 55.57% in generating readable solutions, confirming its feasibility and effectiveness.

Key words: electrical problems, sequence-to-formula tree model, relations extraction, readable solver, Graph Neural Network(GNN)

菅朋朋, 刘浩宇, 闫鸣, 王彦丽, 杨阳蕊, 刘雪梅. 基于序列公式树模型的电学问题解答方法[J]. 计算机工程, 2024, 50(11): 399-408.

JIAN Pengpeng, LIU Haoyu, YAN Ming, WANG Yanli, YANG Yangrui, LIU Xuemei. Solving Electrical Text Problems Based on Sequence to Formula Tree Model[J]. Computer Engineering, 2024, 50(11): 399-408.

https://www.ecice06.com/CN/Y2024/V50/I11/399

图/表 13

图1 Seq2FT模型的整体结构与数据传输

Fig.1 Overall structure and data transmission of Seq2FT model

图2 问题文本预标注过程

Fig.2 Pre-annotation process of problem text

图3 直陈关系组获取过程

Fig.3 The acquisition process of direct relation group

图4 Seq2FT模型的工作流程

Fig.4 Work procedure of Seq2FT model

图5 Seq2Tree模型的求解式结构

Fig.5 The solver structure of Seq2Tree model

图6 Seq2FT模型的求解式结构

Fig.6 The solver structure of Seq2FT model

图7 模型的损失函数对比

Fig.7 Loss function comparison of models

参考文献 31

1	WANG Y, LIU X J, SHI S M. Deep neural solver for math word problems[C]//Proceedings of 2017 Conference on Empirical Methods in Natural Language Processing. Stroudsburg, USA: Association for Computational Linguistics, 2017: 845-854.
2	WANG L, WANG Y, CAI D, et al. Translating a math word problem to an expression tree[EB/OL]. [2023-08-10]. http://arxiv.org/abs/1811.05632.
3	XIE Z, SUN S. A goal-driven tree-structured neural model for math word problems[C]//Proceedings of the 28th International Joint Conference on Artificial Intelligence. Washington D. C., USA: IEEE Press, 2019: 384-396.
4	WU Q Z, ZHANG Q, FU J L, et al. A knowledge-aware sequence-to-tree network for math word problem solving[C]//Proceedings of International Conference on Empirical Methods in Natural Language Processing. Stroudsburg, USA: Association for Computational Linguistics, 2020: 7137-7146.
5	GUO J Y, ZHAO J, YU Z T, et al. Research on semantic label extraction of domain entity relation based on CRF and rules[C]//Proceedings of Asia-Pacific Web Conference. Berlin, Germany: Springer, 2012: 154-162.
6	NAJAFI S, FYSHE A. Weakly-supervised questions for zero-shot relation extraction[EB/OL]. [2023-08-10]. http://arxiv.org/abs/2301.09640.
7	YAN Z, JIA Z, TU K. An empirical study of pipeline vs. joint approaches to entity and relation extraction[EB/OL]. [2023-08-10]. https://arxiv.org/pdf/2312.03025.
8	HONG Y N, LI Q, GONG R, et al. SMART: a situation model for algebra story problems via attributed grammar. Artificial Intelligence, 2021, 35 (14): 13009- 13017.
9	HUANG D Q, SHI S M, LIN C Y, et al. Learning fine-grained expressions to solve math word problems[C]//Proceedings of International Conference on Empirical Methods in Natural Language Processing. Stroudsburg, USA: Association for Computational Linguistics, 2017: 805-814.
10	LIANG C C, HSU K Y, HUANG C T, et al. A tag-based statistical English math word problem solver with understanding, reasoning and explanation[C]//Proceedings of the 25th International Joint Conference on Artificial Intelligence. New York, USA: ACM Press, 2016: 4254-4255.
11	QIN J, LIN L, LIANG X, et al. Semantically-aligned universal tree-structured solver for math word problems[EB/OL]. [2023-08-10]. http://arxiv.org/abs/2010.06823.
12	WU Q Z, ZHANG Q, WEI Z Y, et al. Math word problem solving with explicit numerical values[C]//Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing. Stroudsburg, USA: Association for Computational Linguistics, 2021: 5859-5869.
13	GUO F C, JIAN P P, WANG Y L, et al. A framework of cross-modal learning for solving geometry problems[C]//Proceedings of IEEE International Conference on Engineering, Technology & Education. Washington D. C., USA: IEEE Press, 2021: 506-512.
14	JIAN P P, GUO F C, WANG Y L, et al. Solving geometry problems via feature learning and contrastive learning of multimodal data. Computer Modeling in Engineering & Sciences, 2023, 136 (2): 1707- 1728.
15	GUO F C, JIAN P P. A graph convolutional network feature learning framework for interpretable geometry problem solving[C]//Proceedings of International Conference on Intelligent Education and Intelligent Research. Washington D. C., USA: IEEE Press, 2022: 59-64.
16	LIN X, HUANG Z Y, ZHAO H K, et al. HMS: a hierarchical solver with dependency-enhanced understanding for math word problem. Artificial Intelligence, 2021, 35 (5): 4232- 4240.
17	ZHANG W, SHEN Y, MA Y, et al. Multi-view reasoning: consistent contrastive learning for math word problem[EB/OL]. [2023-08-10]. http://arxiv.org/abs/2210.11694.
18	JIE Z, LI J, LU W. Learning to reason deductively: math word problem solving as complex relation extraction[EB/OL]. [2023-08-10]. http://arxiv.org/abs/2203.10316.
19	LI A, JIANG X, LIU B, et al. Tackling math word problems with fine-to-coarse abstracting and reasoning[EB/OL]. [2023-08-10]. http://arxiv.org/abs/2205.08274.
20	FALDU K, SHETH A, KIKANI P, et al. MMTM: multi-tasking multi-decoder transformer for math word problems[EB/OL]. [2023-08-10]. http://arxiv.org/abs/2206.01268.
21	JIAN P P, SUN C, YU X G, et al. An end-to-end algorithm for solving circuit problems. International Journal of Pattern Recognition and Artificial Intelligence, 2019, 33 (7): 1940004. doi: 10.1142/S0218001419400044
22	菅朋朋. 机器解答电路题目方法研究[D]. 武汉: 华中师范大学, 2019.
	JIAN P P. Research on the method of solving circuit problems by machine[D]. Wuhan: Central China Normal University, 2019. (in Chinese)
23	HE B, YU X G, JIAN P P, et al. A relation based algorithm for solving direct current circuit problems. Applied Intelligence, 2020, 50 (7): 2293- 2309. doi: 10.1007/s10489-020-01667-7
24	YU X G, JIAN P P, WANG M S, et al. Extraction of implicit quantity relations for arithmetic word problems in Chinese[C]//Proceedings of International Conference on Educational Innovation through Technology. Washington D. C., USA: IEEE Press, 2016: 242-245.
25	夏盟. 基于机器解答的物理电路学习资源标注与呈现研究[D]. 武汉: 华中师范大学, 2019.
	XIA M. Research on labeling and presentation of physical circuit learning resources based on machine solution[D]. Wuhan: Central China Normal University, 2019. (in Chinese)
26	王帅. 面向机器解答的电路元器件识别研究[D]. 武汉: 华中师范大学, 2020.
	WANG S. Research on circuit component recognition for machine solution[D]. Wuhan: Central China Normal University, 2020. (in Chinese).
27	张淳. 面向智能导学的初中物理电路题可读解答生成研究[D]. 武汉: 华中师范大学, 2020.
	ZHANG C. Research on readable solution generation of junior high school physical circuit problems for intelligent tutoring[D]. Wuhan: Central China Normal University, 2020. (in Chinese)
28	CAO Y X, HONG F, LI H W, et al. A bottom-up DAG structure extraction model for math word problems. Artificial Intelligence, 2021, 35 (1): 39- 46.
29	LI S, WU L, FENG S, et al. Graph-to-tree neural networks for learning structured input-output translation with applications to semantic parsing and math word problem[EB/OL]. [2023-08-10]. http://arxiv.org/abs/2004.13781.
30	JIAN P, GUO F, PAN C, et al. Interpretable geometry problem solving using improved RetinaNet and graph convolutional network. Electronics, 2023, 12 (22): 4578. doi: 10.3390/electronics12224578
31	SHAKARIAN P, KOYYALAMUDI A, NGU N, et al. An independent evaluation of ChatGPT on mathematical word problems[EB/OL]. [2023-08-10]. http://arxiv.org/abs/2302.13814.

[1]	李泽霖, 吕兆峰, 陈富强, 李克. 基于多跳信息融合的实体对齐模型[J]. 计算机工程, 2024, 50(9): 142-152.
[2]	何杏宇, 周易歆, 罗东旭, 杨桂松. 基于图神经网络和多主体评价的教学资源推荐[J]. 计算机工程, 2024, 50(7): 13-22.
[3]	张正康, 杨丹, 聂铁铮, 寇月. 基于图结构聚类的自监督学习疾病诊断方法[J]. 计算机工程, 2024, 50(7): 360-371.
[4]	孙文洁, 李宗民, 孙浩淼. 基于图神经网络的多智能体强化学习值函数分解方法[J]. 计算机工程, 2024, 50(5): 62-70.
[5]	游奔, 李晓红, 姚锦, 冯绍杰. 基于多粒度图与注意力机制的半监督短文本分类[J]. 计算机工程, 2024, 50(5): 83-90.
[6]	张宝鑫, 杨丹, 聂铁铮, 寇月. 基于自监督的多视角图协同过滤推荐方法[J]. 计算机工程, 2024, 50(5): 100-110.
[7]	吴正洋, 张广涛, 黄立, 汤庸. 基于异质图嵌入和会话交互的课程推荐模型[J]. 计算机工程, 2024, 50(4): 95-103.
[8]	王华珍, 许泽, 孙悦, 丘斌, 陈坚, 邱强斌. 基于事件演化图的多标记事件预测模型[J]. 计算机工程, 2024, 50(4): 132-140.
[9]	王俊, 赖会霞, 万玥, 张仕. 基于角度的图神经网络高维数据异常检测方法[J]. 计算机工程, 2024, 50(3): 156-165.
[10]	王柏涵, 姜晓燕, 范柳伊. 基于深度监督隐空间构建的语义分割改进方法[J]. 计算机工程, 2024, 50(3): 191-199.
[11]	李伟健, 胡慧君. 基于潜在特征增强网络的视频描述生成方法[J]. 计算机工程, 2024, 50(2): 266-272.
[12]	潘伟, 黄瑞章, 任丽娜, 薛菁菁. 基于自适应结构学习的深度文本聚类[J]. 计算机工程, 2024, 50(11): 89-97.
[13]	沈学利, 刘士枫. 基于图边缘特征注意力的入侵检测模型[J]. 计算机工程, 2024, 50(11): 236-245.
[14]	顾宇衡, 潘嘉诚, 钱江波, 董一鸿. 阿尔茨海默病的图神经网络分类方法研究进展[J]. 计算机工程, 2024, 50(10): 35-50.
[15]	吴志强, 解庆, 李琳, 刘永坚. 基于多模态融合的图神经网络推荐算法[J]. 计算机工程, 2024, 50(1): 91-100.

选择文件类型/文献管理软件名称

选择包含的内容