Parsing the semantic structure of Indonesian math word problems using the recursive neural network

Abstract

Math word problems play an important role for the development of artificial intelligent. This is because solving word problems involves the development of a system that can understand natural language. Designing a system for solving math word problems requires a mechanism for decomposing a text into segments of text to be translated into math operation. The segments are categorized through the process of parsing the semantic structure of the word problems to obtain segments whose meanings refer to math operation. A number of current proposed methods are suitable to be applied to English math word problems and have never been applied to Indonesian math word problems. The impact is that the segments produced are not necessarily in line with the sequences of operations appropriate with the meaning of the story. This study proposed the use of Recursive Neural Network (RNN) as a parser of semantic structure of Indonesian math word problems. The testing of the parser was carried out on the math word problems taken from the Elementary School’s Electronic School Book (BSE) published by the Book Center of the Ministry of Education and Culture. The result of the testing showed that the final accuracy was 86.4%.