Analyzing TF-IDF and Word Embedding for Implementing Automation in Job Interview Grading
Annalisa Wahyu Romadon, Kemas Muslim Lhaksmana, Isman Kurniawan, Donni Richasdy · 2020
Selecting the best talents from a large number of job applicants is challenging, especially for big companies that usually receive tens of thousands of applicants for every job opening. One of the most costly and time-consuming applicant selection stages is the interview process, since it usually performs face to face meetings and involves third parties to do the interviews and analyze the result. To this end, Human Capital Directorate at Telkom Indonesia adopts AI technology to automate some stages of job applicant selection to reduce manual process and third-party involvement. In this paper, we investigate appropriate feature extraction methods to automate job interview grading for reducing bias and human errors. TFIDF, one of the most popular feature extractions, is compared with word embedding to find the optimal method and parameters in classifying interview verbatims with ANN classifier. Based on the test results, the average accuracy for TFIDF outperforms word embedding by 85.22% against 74.88%, respectively. Therefore, for the case of job interview grading using our dataset, TF-IDF performs better to reduce the number of dimensions.