Automatic Assessment of Answers to Mathematics Stories Question Based on Tree Matching and Random Forest

Math word problems can be solved with a good understanding of language, correct translation, and use of proper operations. However, elementary school students will only get the correct answer if the result is done correctly. Then a numeracy competency detection system is needed through the answer of math story questions. This study aims to build a system of checking student answers in stages. The main contribution is the technique for comparing the trees from two multimodal input and assess student answer automatically. System extracted operand from math story sentences and classified operator using random forest to generate the key then convert it to the tree. We use OCR library to extract text from student’s answer image and identify operand, operator, and result to build student’s answer tree. A tree matching is applied to compare the similarity of trees for automatic assessment. The dataset used in this research is 500 questions, 300 data for training, and 200 data for testing. There are two categories of questions, single and mixed operator, with five class namely addition, subtraction, multiplication, division, and mixed. Based on the experiment, the accuracy of classification for mixed operator is 68.8%, whether for single operator is 84.31%. For tree matching, we achieved 78.12%