학술논문
대학의 AI 기반 맞춤형 강의 추천 시스템 개발 및 실제 적용 사례 연구: K 대학을 중심으로
Development and Application of an AI-Powered Adaptive Course Recommender System in Higher Education: An Example from K University
Development and Application of an AI-Powered Adaptive Course Recommender System in Higher Education: An Example from K University
Document Type
Article
Author
이진숙 / Jinsook Lee; 문기범 / Kibum Moon; 한수연 / Suyeon Han; 이수강 / Sukang Lee; 권혜정 / Hyejung Kwon; 한재호 / Jaeho Han; 김규태 / Gyutae Kim
Source
교육공학연구 / Journal of Educational Technology. Jun 30, 2021 37(2):267
Subject
Language
Korean
ISSN
1225-424X
Abstract
본 연구는 대학혁신을 위한 인공지능 기반의 적응형 학습인 AI 기반 교양 강의 추천 시스템을 래피드 프로토타입 모형에 기반하여 개발하고, 실제 교내 포털 시스템에 적용하여 이용 결과를 분석하는 것에 목적을 두었다. 해당 서비스는 2020년 7월 교내 포털 시스템에 적용하였다. 추천 기능에 이용된 알고리즘은 사용자 기반 협업 필터링과 수강 이력 기반 통계 알고리즘을 이용하였으며, 각 모델당 21개의 교양 강의를 추천하였다. 서비스 만족도 설문조사를 진행한 결과, 782명의 응답을 수집하였고 협업 필터링 알고리즘보다 통계 기반 알고리즘의 만족도가 유의미하게 높은 것을 확인하였다. 그러나 실제 사후 추적 조사 결과, 2020년 2학기 희망 강의로 등록된 강의 내역과 실제 수강 내역에서 추천된 강의를 분석했을 때 협업 필터링 알고리즘의 Recall@21이 각각 약 37%와 43%로 통계 기반 알고리즘의 결과인 18%와 14%에 비해 높은 것으로 나타났다. 또한, 학생들은 교양 강의를 선택할 때 흥미 및 관심사를 가장 우선순위로 고려하였으며, 강의 제목의 모호함 때문에 강의에 대한 키워드가 가장 필요한 정보라고 응답하였다. 더불어 설문 응답자들은 원하는 강의와 원하는 수업 방식을 추천 결과에 직접 반영하고자 하는 요구를 확인하였다. 본 연구가 국내 대학 교육 실정에 맞는 인공지능 기반의 맞춤형 강의 추천 시스템을 개발하고 학습자에게 맞춤형 교육 정보를 제공하고자 할 때 기초자료로 기여할 수 있기를 기대한다.
This paper outlines the development process of an AI-based elective course recommendation system on the basis of rapid prototype methodology (RP), from algorithm modeling to development of a user interface and follow-up survey. The algorithms used to produce recommendations employed either user-based collaborative filtering or a class history-based statistical model, incorporating students’ course ratings and course enrollment history data. The system was implemented on the campus portal website in July 2020, and a satisfaction survey was conducted. Our results, based on 782 responses, demonstrated that the statistical-based model had significantly higher satisfaction than the collaborative filtering model. However, a follow-up survey based on course wish list and course registration data found that Recall@21 for the collaborative filtering model was about 37% and 43%, respectively, compared with 18% and 14%, respectively, for the statistical-based model. Thus, we found a difference between satisfaction with the recommended list and actual course behavior. In their responses, students regarded their academic interests as the top priority when choosing elective courses, and noted that keywords, capable of fully describing the lectures, were vital information due to ambiguous course titles. This study is expected to contribute to the further development and real application of AI-based recommendation systems in Korean higher education institutions.
This paper outlines the development process of an AI-based elective course recommendation system on the basis of rapid prototype methodology (RP), from algorithm modeling to development of a user interface and follow-up survey. The algorithms used to produce recommendations employed either user-based collaborative filtering or a class history-based statistical model, incorporating students’ course ratings and course enrollment history data. The system was implemented on the campus portal website in July 2020, and a satisfaction survey was conducted. Our results, based on 782 responses, demonstrated that the statistical-based model had significantly higher satisfaction than the collaborative filtering model. However, a follow-up survey based on course wish list and course registration data found that Recall@21 for the collaborative filtering model was about 37% and 43%, respectively, compared with 18% and 14%, respectively, for the statistical-based model. Thus, we found a difference between satisfaction with the recommended list and actual course behavior. In their responses, students regarded their academic interests as the top priority when choosing elective courses, and noted that keywords, capable of fully describing the lectures, were vital information due to ambiguous course titles. This study is expected to contribute to the further development and real application of AI-based recommendation systems in Korean higher education institutions.