In online education scenario, recommending exercises for students is an attractive research topic. In this paper, we propose a new hybrid recommendation model that combines deep collaborative filtering (DeepCF) component with wide linear component. The former incorporates stacked denoising auto-encoder(SDAE) into matrix factorization and the latter is general linear component. In DeepCF component, we employ SDAE to learn low dimension latent feature of a student's feature and an item's feature and use matrix factorization method to predict the rating that a student rates an item. In wide linear model, we incorporate some meta properties of an item, such as difficulty, type and knowledge components(KCs). The two components are combined by linear approach. We use negative sampling method to generate the training dataset. An item is corrupted by Gaussian noise and is feed into the SDAE net ,which consists of encoder and decoder with multiple layers. We use tightly couple model to combine SDAE model and collaborative filter model. Experimental results show that the proposed model achieves a 10% relative improvement in AUC metric compared to the traditional collaborative filter method.

Abstract. In recent years, online education has been advancing significantly. However there is a major challenge how to evaluate style of teachers and state of student learning. In this paper, we propose a novel method that combines speaker diarization, speaker recognition, feature selection to classify style of teachers and state of students learning based on audio data. We train speaker recognition model and learn embedding vector of teachers or students on online platform. We select 25 acoustic features and statistical features from audio recordings and train classification model to classify style of teachers and state of students' learning jointly. Experimental results show that the task of classifying style of teachers achieves 71.25% precision and precision of classifying state of students' learning is 83.71%.