Adaptive learning path recommender approach using auxiliary learning objects

• Published in: Computers and education 2020-04, Vol.147, p.103777, Article 103777
• Main Authors: Nabizadeh, Amir Hossein, Gonçalves, Daniel, Gama, Sandra, Jorge, Joaquim, Rafsanjani, Hamed N.
• Format: Article
• Language: English
• Subjects: Auxiliary learning objects; e-learning; Item Response Theory (IRT); Learning path; Long Term Goal Recommenders (LTRS); Matrix Factorization (MF); Recommendation Systems (RS)
• ISSN: 0360-1315; 1873-782X
• DOI: 10.1016/j.compedu.2019.103777

In e-learning, one of the main difficulties is recommending learning materials that users can complete on time. It becomes more challenging when users cannot devote enough time to learn the entire course. In this paper, we describe two approaches to maximize users’ scores for a course while satisfying their time constraints. These approaches recommend successful paths based on the available time and knowledge background of users. We first briefly explain a method that has a similar goal to our method, and highlight its drawbacks. We then describe our proposal, which works based on a two-layered course graph (lesson and Learning Object (LO) layers; a lesson includes a few LO). Initially, our method uses the Depth First Search algorithm (DFS) to find all lesson sequences in the graph that start by a lesson (opted by a user). It then assigns LO for lessons of paths and estimates their score and time. Finally, a path that satisfies the user’s limited time while maximizing his/her score is recommended lesson by lesson. During a path recommendation, if the user could not get the estimated score from a lesson, our method recommends auxiliary LO for that lesson. To evaluate our method, we first assessed the quality of our estimation methods and then evaluate our recommender in a live environment. Results show that our estimation methods outperformed the ones in the literature. Results also present within the same amount of time, the users of our recommender proceeded more on the course than the users of another e-learning system. •This paper includes an extensive and a comprehensive related work.•We introduce 3 methods to estimate learning time and score for paths, which are: Clust.Mean, Clust.Median, MF.Predict. We also compare their performance with six other methods that are introduced in the literature.•In this paper, we point out the main drawbacks of a method in the literature that has a similar goal to our method.•Our proposal uses a two-layered course graph, which results in defining all relations among the learning objects (LO) of a course and also to have a control environment for the path recommendation.•Our method is an adaptive one, which enables the recommender to adapt a path regarding a user’s progress.•The proposed method recommends auxiliary LO. These LO are not in the initial generated path for a user, and they will be recommended whenever a user could not learn a lesson properly (not obtaining the expected score).•In this paper, we conduct an A