APPLICATION OF SEMI-SUPERVISED LEARNING FOR JOB APPLICANT COMPETENCY CLASSIFICATION

In an effort to gather high-quality human resources or meet qualifications, an organization needs to go through a series of stages in the candidate recruitment process. One of these stages is the interview process. The interview process can be optimized with the help of machine learning technology....

Full description

Saved in:
Bibliographic Details
Main Author: Alexander Wen, Steven
Format: Final Project
Language:Indonesia
Online Access:https://digilib.itb.ac.id/gdl/view/85482
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Institut Teknologi Bandung
Language: Indonesia
Description
Summary:In an effort to gather high-quality human resources or meet qualifications, an organization needs to go through a series of stages in the candidate recruitment process. One of these stages is the interview process. The interview process can be optimized with the help of machine learning technology. By utilizing the answers from candidate interviews, these answers can later be classified into competency levels present in the competency dictionary that has been determined by the organization. The candidate competency classification model was built using a pre-trained model, which was then fine-tuned using semi-supervised learning methods, namely pseudo-label and ladder network gamma model. This semi-supervised learning method allows the model to perform continuous learning. This Candidate Competency Classification Model is part of the subsystem. For communication with other subsystems, a classification model API is provided. The model trained using the pseudo-label and ladder network gamma model methods was evaluated using a dataset of 36 labeled data points and 30 unlabeled data points, which were then combined with 28 data points used as the training set and 8 data points as the validation set. The model trained using the pseudo-label method achieved a best accuracy of eighty-seven point five percent, and the model trained using the ladder network gamma model method achieved the highest accuracy of one hundred percent during the validation test. The Indonesian language-based Competency Assessment subsystem was successfully developed and integrated into the entire application subsystem. The Indonesian language-based competency assessment model also supports continuous learning.