Video Recommender System using textual data: Its application on LMS and Serendipity evaluation

Abstract

Learning Management System (LMS) has become a tool common to teachers and students in this day and age. Students focus on the materials posted by their teachers, be they slides, lectures, class notes, or other documents. But with the massive pool of knowledge on the internet, students should be able to learn from these external resources if they can identify the useful ones. In this work, we propose an algorithm and an application that automatically search and select videos from YouTube that are relevant to the materials posted on Learning Management System (LMS). Our application is called Video Recommender System (VRS). It contains two parts; API and Document Processing. We utilized the Term Frequency-Inverse Document Frequency (TF-IDF) to map the documents and videos based on their metadata. The evaluation method used in this study is based on Serendipity which is designed to measure the preferable, unexpected or good-surprise outcomes. Finally, we compare the result of our algorithm to one from direct search using material titles. A/B testing revealed that our recommendations outperform the traditional search and the average Serendipity scores for VRS is 0.196 greater than that of the title search with significance.