Matrix factorization is the most popular approach to solving prediction problems. However, in the recent years multiple relationships amongst the entities have been exploited in order to improvise the state-of-the-art systems leading to a multi relational matrix factorization (MRMF) model. MRMF deals with factorization of multiple relationships existing between the main entities of the target relation and their metadata. A further improvement to MRMF is the Weighted Multi Relational Matrix Factorization (WMRMF) which treats the main relation for the prediction with more importance than the other relations. In this paper, we propose to enhance the prediction accuracy of the existing models by personalizing it based on student knowledge and task difficulty. We enhance the WMRMF model by incorporating the student and task bias for prediction in multi-relational models. Empirically we have shown using over five hundred thousand records from Knowledge Discovery dataset provided by Data Mining and Knowledge Discovery competition that the proposed approach attains a much higher accuracy and lower error(Root Mean Square Error and Mean Absolute Error) compared to the existing models.
Prof. Prema Nedungadi and Smruthy, T. K., “Personalized Multi-relational Matrix Factorization Model for Predicting Student Performance”, in Intelligent Systems Technologies and Applications: Volume 1, S. Berretti, Thampi, S. M., and Srivastava, P. Ranjan, Eds. Cham: Springer International Publishing, 2016, pp. 163–172.