INCREASING THE RELIABILITY OF PREDICTING EDUCATIONAL TRAJECTORIES BY A RECOMMENDATION SYSTEM THAT UNITES ACADEMIC AND MOTIVATIONAL PROFILES

Abstract

The present study is due to the fact that traditional recommendation systems demonstrate not very high course selection accuracy. The focus of the present study was the construction and empirical validation of the use of the proposed recommender system. The study intended to build a recommendation system and test it for predicting the educational trajectories of IT majors of VKTU.

To address the research problem, matrix factorization was applied in modeling the academic profile of the learners. A motivational vector building on the RIASEC psychometric modeling was created in parallel. The results of the two predictive components were combined through a linear weighted combination, whereas the coefficient in this combination was optimized by the Grid Search method. Accuracy and stability evaluation was performed using five-fold cross-validation and key metrics.

The found coefficient value of α=0.58 gave the optimal parameters of our system. At this coefficient, the highest prediction accuracy and the lowest error were obtained. Our model demonstrated significant results stability. The study confirms that the combination of academic and motivational profiles built into our system can be one of the components for developing open and reliable recommender systems.