Graph2Tree 모델을 이용한 한국어 수학 문장제 문제 풀이 = A Graph2Tree Model for Solving Korean Math Word Problems

다국어 초록 (Multilingual Abstract)

This paper builds its own data set of eight types of Korean math word problems and presents a Ko-Graph2Tree model, an automatic solution model for Korean math word problems based on Graph2Tree model not previously presented. The recently released Graph2Tree model is a graph-to-tree learning based model that shows better performance than existing natural language processing models for automatic solving English math word problems. Using two types of graphs reflecting the relationship and order between numbers in the problem text, that is, mathematical relations, in solution generation, the model showed improved performance compared to existing tree-based models. As a result of measuring performance after learning with a self-produced Korean math word problem dataset, the transformer model with sequence-to-sequence structure showed an accuracy of 42.3%, whereas the Ko-Graph2Tree model showed an accuracy of 68.3%, resulting in 26.0%p higher performance.

국문 초록 (Abstract)

본 논문은 8개 유형으로 이루어진 한국어 수학 문장제 문제 데이터셋을 자체적으로 구축하여 이를 기반으로 기존에 제시되지 않았던 Graph2Tree 모델 기반 한국어 수학 문장제 문제 자동 풀이 모델인 Ko-Graph2Tree 모델을 제시한다. 최근 공개된 Graph2Tree 모델은 영어 수학 문장제 문제 자동 풀이에 기존의 자연어 처리 모델들보다 뛰어난 성능을 보인 모델이다. 해당 모델은 문제 텍스트 내의 숫자 간 관계성 및 순서, 즉 수학적 관계를 반영한 두 가지의 그래프를 풀이 생성에 사용함으로써 기존의 트리 기반 모델들보다 향상된 성능을 보인다. 자체 제작한 한국어 수학 문장제 문제 데이터셋으로 학습시킨 후 성능을 측정한 결과, 시퀀스 투 시퀀스 구조의 트랜스포머 모델은 정확도가 42.3%, 본 논문이 제시한 Ko-Graph2Tree 모델은 정확도가 68.3%로 26.0%p 더 높은 성능을 보였다.

참고문헌 (Reference)

1 G. Lee, "embedding resources"

2 Q. Liu, "Treestructured Decoding for Solving Math Word Problems" 2370-2379, 2019

3 L. Wang, "Translating a Math Word Problem to a Expression Tree" 1064-1069, 2018

4 F. Scarselli, "The Graph Neural Network Model" 20 (20): 61-80, 2009

5 L. Wang, "Template-Based Math Word Problem Solvers with Recursive Neural Networks" 33 (33): 7144-7151, 2019

6 I. Sutskever, "Sequence to Sequence Learning with Neural Networks" 3104-3112, 2014

7 T. N. Kipf, "Semi-Supervised Classification with Graph Convolutional Networks" 2017

8 T. -R. Chiang, "Semantically-Aligned Equation Generation for Solving and Reasoning Math Word Problems" 1 : 2656-2668, 2019

9 K. H. Cho, "On the Properties of Neural Machine Translation : Encoder-Decoder Approaches" 103-111, 2014

10 D. Bahdanau, "Neural Machine Translation by Jointly Learning to Align and Translate" 2015

11 L. Dong, "Language to Logical Form with Neural Attention" 1 : 33-43, 2016

12 E. L. Park, "KoNLPy : Korean natural language processing in Python" 133-136, 2014

13 J. Park., "KoELECTRA: Pretrained ELECTRA Model for Korean"

14 J. Zhang, "Graph-to-tree Learning for Solving Math Word Problems" 3928-3937, 2020

15 S. Li, "Graph-to-Tree Neural Networks for Learning Structured Input-Output Translation with Applications to Semantic Parsing and Math Word Problem" 2841-2852, 2020

16 K. Clark, "ELECTRA : Pre-training Text Encoders as Discriminators Rather Than Generators" 2020

17 Y. Wang, "Deep Neural Solver for Math Word Problems" 845-854, 2017

18 J. Devlin, "BERT : Pre-training of Deep Bidirectional Transformers for Language Understanding" 1 : 4171-4186, 2019

19 A. Vaswani, "Attention is All you Need" 5998-6008, 2017

20 Z. Xie, "A Goal-Driven Tree-Structured Neural Model for Math Word Problems" 5299-5305, 2019

21 Z. Wu, "A Comprehensive Survey on Graph Neural Networks" 32 (32): 4-24, 2021