Implementation of TF-IDF Method and Support Vector Machine Algorithm for Job Applicants Text Classification

Abstract

Tens of thousands of people are applying for job in PT. Telkom each year. The goal of the recruitment process is to get new employees which can fit PT. Telkom's working culture. Due to the high number of applicants, the recruitment process takes a lot of time and affecting higher cost to spend. We're proposing a popular combination of Term Frequency-Inverse Document Frequency (TF-IDF) as the extraction method and Support Vector Machine (SVM) as the classifier to filter the applicants' interview text. SVM generally produces better accuracy in text classification compared to Random Forest or K-Nearest Neighbors (KNN) algorithm. However, TF-IDF has several developments to improve its flaws, one of them is Term Frequency-Relevance Frequency (TF-RF). As a comparison, in this study we use three extraction methods: TF only (without IDF), TF-IDF, and TF-RF. We use interview texts from PT. Telkom as the data source. The results of combination SVM with TF-IDF can produce 86.31\% of accuracy, with TF only can produce 85.06\%, and with TF-RF can produce 83.61\% of accuracy. The results show extracting method TF-IDF can still outperform TF-RF in term of accuracy.