On Supporting Game-Based Learning via Recommendations

Abstract

Over the last two decades, game-based learning has gained increasing popularity. In today's world, teachers are expected to utilize technological tools such as digital games as learning aids. Despite the multitude of studies examining the benefits of game-based learning, finding the most convenient game for a particular teaching purpose can be a challenging task given the vast number of similar games that are available on the market. With this study, we aim to provide teachers with a recommendation system that will assist them in selecting appropriate games from all the web-based game materials available. A key theoretical premise behind this work is to examine teaching from the perspective of teachers to develop their ability to teach. The purpose of this study is to develop a recommendation system that will assist teachers in selecting educational games based on the subjects they teach, that will be both personalized and use the experience of other researchers at the same time. We propose a system that utilizes the latest developments in signal processing and machine learning, specifically the tensor completion method. This is a machine learning technique from the family of collaborative filtering methods that fills in missing values in a dataset by analyzing its existing patterns. According to our knowledge, this is the first study to modify a collaborative filtering approach to develop a recommendation system for game-based learning. Whenever a teacher requires a recommendation, the method leverages other users' ratings on games, while also considering the teacher's previous experience with the system. Accordingly, the system selects the best games for users based on their previous preferences as well as the experiences of other users. It appears that this system has a reasonable chance of working properly without excessive training samples, which could not be achieved through other advanced machine-learning techniques. The experimental section demonstrates the potential for such a proof-of-concept technique, which uses tensor completion; even with a small number of collected data, performance starts to increase in suggesting games to the specific user, and this trend is promising for big data.