<![CDATA[Math word problems act as a test bed to design an intelligent system. An approach is needed to identify arithmetic operations including addition, substraction, multiplication and division. Template-based approaches have addressed this problem. However, the template-based approach is less efficient because it requires the process of building a template repository that have to cover a wide variety of story implied by math words. The template-based approach is potentially sub-optimal when solving story problems that have not been covered yet by templates. The proposed approach resolves this by using Recursive Neural Network and Support Vector Machine. Recursive neural network is used as an encoder that can generate semantic vectors of math word problems. Then, this vector becomes as an input for a Support Vector Machine-based classifier. Tests were conducted on a dataset collected manually from Kemdikbud’ electronic school books. The results showed that the proposed approach does not require the formation of templates, thereby reducing human involvement. In addition, the use of Recursive Neural Network reduces feature engineering making it more efficient. Experimental results by applying k-fold cross validation show that the proposed approach has an accuracy of 81% and a precision of 66%]]>
Copyrights © 2024
