Learner Modeling in Educational Games Based on Fuzzy Logic and Gameplay Data

Introduction

Recently, we have witnessed the emergence of educational games as learning tools. In fact, today’s students grow up in the video games era; playing a video game is a part of the daily routine of the majority of them. As a result, schools started resorting to them more than ever to ensure deeper engagement and to provide adaptive and personalized learning. Many researchers have proven the efficiency of educational games as learning tools (Gee, 2003; Chizary & Farhangi, 2017; Castronovo, Van Meter & Messner, 2018; Ge & Ifenthaler, 2018; Collins, 2011). In addition to their immediate and incidental feedback to learners, educational games also allow students to progress at their own pace and grant them the opportunity to explore, by trying new things and taking risks in a safe place without being openly judged or ranked (Gee, 2003). They also allow individualized learning where the learning content, activities, materials, and pace of learning are adjusted to each learner; which leads to increased motivation, greater retention of knowledge and deeper understanding. To provide adaptive learning paths, it is essential to have an adequate and detailed representation of the learners’ models including their needs, traits, preferences and proficiency. Bull (2004) defined the learner model as follows “the learner model is a model of the knowledge, difficulties and misconceptions of the individual. As a student learns the target material, the data in the learner model about their understanding is updated to reflect their current beliefs”.

In learning systems, many studies (Gorgun, Turker, Ozan & Heller, 2005; Sani, Mohammadian & Hoseini, 2012; Vagale & Niedrite, 2012) have agreed that the learner’s model should contain the knowledge level, interactions, interests and learning style. Learning style stands for the way a student prefers to learn. Taking into account learning styles while designing learning activities, increases learners’ cognitive motivation and helps to create personalized learning (Filippidis & Tsoukalas, 2009; Graf, Liu & Kinshuk, 2010; Hauptman & Cohen, 2011; Hwang, Sung, Hung & Huang, 2013; Hamza & Tlili, 2018; Wan, San, & Omar 2018; Samia & Amirat, 2018; Wouters, & van der Meulen, 2020). Good educational games also adjust the gameplay activities and difficulty to match the player’s playing style. Playing styles determine the player’s preferences in terms of gameplay. Adapting the gameplay to the players’ playing style helps to improve their performances in the game (Charles, Kerr, & McNeill 2005; Wong, Kim, Han & Jung, 2009; Shaker, Yannakakis & Togelius, 2010).

Many techniques can be used to model learners in educational games. Khenissi, Essalmi, Jemni & Kinshuk, 2015, presented a literature review where they identified the different existing learners modeling methods in educational games. Those methods can be categorized in two groups, explicit and implicit modeling. Explicit modeling extracts information about the learner in an obvious and explicit way using procedures like questionnaires (Moreno Ger, Sancho Thomas, Martínez Ortiz, Sierra & Fernández Manjon, 2007; Huang 2011; Fu, Su & Yu 2009; Pourabdollahian, Taisch & Kerga, 2012) or software and hardware equipment to determine learner’s behavior by identifying their gestures, body posture and physiological signals (Amershi, Conati & Maclaren, 2006; Rebolledo-Mendez & De Freitas, 2008; Conati & Maclaren, 2009; Peters, Asteriadis & Rebolledo-Mendez, 2009; Sosa Jimenez, Mesa, Rebolledo-Mendez & De Freitas, 2011). Implicit modeling on the other hand collects learner’s data in a hidden and a discrete way. Diverse methods have been used to model the learner implicitly; translating learner’s action is one among them (Lessard, 2012; Stathacopoulou, Samarakou, Grigoriadou & Magoulas, 2004; Conati & Zhao, 2004; Manske & Conati 2005). It translates the actions made by the learner during the game to descriptive information and uses it to model the learner. Monitoring learner’s errors is another technique of modeling learners implicitly (Virvou, Manos & Katsionis, 2003; Rebolledo-Mendez et al. 2008). It helps to model learner’s knowledge, misconceptions and errors. Another method is interpreting interaction traces, it consists of monitoring players’ interactions and generating traces that will be analyzed with the help of a group of experts (Stathacopoulou, Grigoriadou, Samarakou & Mitropoulos, 2007). Players’ conversations can also be used in learners’ modeling. Indeed, useful information can be extracted from players’ responses to questions. For instance, Moreno Ger et al. (2007) resorted to interpreting conversations between the learners and NPCs during the gameplay to identify their learning styles and preferences.