LSTM과 BERT를 활용한 수학 문제 유형 분류

Abstract

Mathematical equations are crucial for expressing complex concepts in science and mathematics, yet their intricate structures and hierarchical relationships present significant challenges for computational interpretation and classification. This study proposes a deep learning model that integrates Long Short-Term Memory (LSTM) networks with Bidirectional Encoder Representations from Transformers (BERT) to enhance equation domain classification. Equations are represented in LaTex format and converted into bigram token sequences for structural pattern recognition by the LSTM encoder, while the masked textual context is processed by the BERT encoder to capture contextual meaning. Based on the two codes, MLP decides on the topic category. Experimental results demonstrate that the combined model outperforms standalone LSTM and BERT models, achieving an accuracy of 0.9196, precision of 0.9211, recall of 0.9197, and F1 score of 0.9200. These findings confirm that integrating structural and contextual information significantly improves equation meaning classification. Future work will explore expanding the dataset for better generalization and changing the neural network models used in the original study.