Identification of Quantity Relationships in Math Story Problems Using Bidirectional Long Short-Term Memory (Bi-LSTM)

Abstract

This research is motivated by the difficulty in automatically identifying and classifying quantity relationships in mathematical word problems due to variations in sentence structure, the use of natural language, and the complexity of the context contained in the text. These problems pose a challenge in the development of natural language processing systems to support mathematical problem understanding. This study aims to identify quantity relationships in mathematical word problems using a deep learning approach based on Bidirectional Long Short-Term Memory (BiLSTM). The research method used is a computational experimental study with stages of data collection, text preprocessing, vocabulary formation, model training, and model performance evaluation. The evaluation was carried out using accuracy, precision, recall, and F1-score metrics to measure the model's ability to classify quantity relationships. The results show that the BiLSTM model is able to identify quantity relationships well, indicated by an accuracy value of 0.9224 and high precision, recall, and F1-score values ​​in each class category. These findings indicate that BiLSTM is effective in understanding the pattern of quantity relationships in mathematical word problems and has the potential to be used as an automated approach in natural language processing-based mathematical text analysis.