A Hybrid Recommender System based on DeepCF and Wide Linear Model for K12 Online Education

Abstract

Compared to traditional classroom education, online education has the advantage of personalized teaching and is able to significantly improve learning efficiency and outcomes. K12 online education refers to online education serving students across 12 grade levels, from primary school to high school. As there are a plethora of courses offered on online education platforms, how to recommend proper courses for students is a serious challenge. In order to resolve this problem, this study proposes a hybrid recommendation model that combines the deep collaborative filtering (DeepCF) model and the wide linear model. Together, in the hybrid model, these models can integrate course features, student features, and side information. The DeepCF model learns low-dimension latent representations for both courses and students and integrates them into matrix factorization to predict ratings. The wide linear model uses a factorization machine to design and select features automatically. The hybrid model can achieve good performance and alleviate the problem of sparse features and cold start. Experimental results demonstrate that compared with the collaborative filter model, the hybrid model achieved a significant improvement with a 12.7% relative increase in AUC metric.