Term Frequency-Inverse Document Frequency Answer Categorization with Support Vector Machine on Automatic Short Essay Grading System with Latent Semantic Analysis for Japanese Language

摘要

In this paper, conducted a research to increase accuracy of Japanese language automatic short essay grading system. Japanese short answers are processed with a supervised machine learning algorithm; Support Vector Machine (SVM) before entering the system that used Latent Semantic Analysis (LSA). The SVM is used to classify short answers topics that minimize error in assessing the essay. TF-IDF process is done as an input to the SVM to weigh every keyword in a sentence. Then, the result will be processed with LSA. LSA uses Singular Value Decomposition (SVD) as the main process and Frobenius Norm as the final calculation from the result of SVD. Using linear kernel in SVM, the accuracy obtained in classifying short answers topics from Japanese-written short answers is 96.36% with 10.0 to 100.0 penalty values and 0.5 training portion. The accuracy score obtained from LSA is as much as 87.15% average with the input of TDM that shows frequency of a word's occurrence.