Our recommendation: Surprisal. A recommender system with information theory for e-learning

Abstract

This paper presents the concept of an automatic recommender system, which employs an information-theoretic approach and is designed for use on e-learning platforms. The proposed approach involves the processing of text and the representation of the required text units in a high-dimensional space. This includes the representation of user responses recorded through an initial user survey as well as course descriptions. In addition to word embeddings, the vectors consist of values that represent the information content of user responses and course descriptions.