Um framework de sistemas de recomendação para uso no ensino não-formal

Abstract

Recommender systems (RS) are present in the most diverse areas. Among them, an area in which RS is gaining more and more relevance is education. Since people are continually learning, not only in a formal way at school/university, one can observe an increasing interest in the search for courses, usually online, that provide some training for the student. We can classify this area of education as non-formal education. Thus, as the student may be learning in different ways and different places, the information that is around him, in his environment, that is, his context, can influence the way the student learns. Firstly, we realize a systematic review of the literature to identify aspects of RS applied in non-formal education. Also, to evaluate if, and how, researchers use the context information in these systems. From this analysis, we propose a recommendation framework that uses the student’s context information. The proposed approach consists of four stages: i) data entry and preparation; ii) filtering and generation of preliminary recommendations; iii) RS hybridization, where we apply the student’s context information to give a closer sense to the student’s reality for the recommendations; and, iv) the exit step, where we provide the suggestions. To evaluate the framework experiments, we conducted an offline manner in a synthetic dataset, constructed with user interaction information with learning objects, and with user check-in information to simulate the context information. We evaluate the results obtained by the framework regarding coverage of the recommendations, the similarity between the recommended learning objects, customization, precision, and recall. As can be seen, the framework presented promising results, with excellent coverage, a balance in the similarity between the recommendations, thus generating some novelty in the recommendations and a recommendation customization rate close to 65%.