Educational Data Mining (EDM) has emerged as a vital field of research, which harnesses the powerof computational techniques to analyze educational data. With the increasing complexity and diversity of educational data, Deep Learning techniques have shown significant advantages in addressingthe challenges associated with analyzing and modeling this data. This survey aims to systematicallyreview the state-of-the-art in EDM with Deep Learning. We begin by providing a brief introduction toEDM and Deep Learning, highlighting their relevance in the context of modern education. Next, wepresent a detailed review of Deep Learning techniques applied in four typical educational scenarios,including knowledge tracing, student behavior detection, performance prediction, and personalizedrecommendation. Furthermore, a comprehensive overview of public datasets and processing tools forEDM is provided. Finally, we point out emerging trends and future directions in this research area.
3.4.3. Reinforcement Learning The application of reinforcement learning in personalized recommendation has high research value and development prospects. Reinforcement learning algorithms can learn the optimal recommendation strategy based on learner feedback to improve the recommendation performance. 3.4.2. Unsupervised Learning In recommender systems, user preferences and behaviors are often incomplete and inaccurate, and tagging data is very difficult and expensive to collect. On this basis, unsupervised learning is a great means to achieve personalized recommendations. Unsupervised learning can extract potential interests and preferences from users historical behaviors and infer the similarity and interest relevance of users through clustering, and feature learning to achieve personalized recommendations.
id: 66476644a35122f905dcdb0889741ebd - page: 11
A cognitive structure enhanced framework for adaptive learning named CSEAL was proposed by Liu et al. , to achieve personalized learning path recommendation. This framework views the learning path as a Markov Decision Process (MDP) and applies actor-critic to identify appropriate learning projects for individual learners. CSEAL comprehensively considers the learners knowledge level and the knowledge structure of the learning project. Experimental results demonstrate that CSEAL can enhance learners efficiency compared with current adaptive learning methods.
id: dc6329ef091d9068d00d1747facb5e21 - page: 11
Bharadhwaj et al. introduced a hybrid RecGAN model based on GAN and RNN. The generator and discriminator are both constructed with GRU-Based RNN. The generator is then allowed to play a Mini-Max game with the discriminator, i.e., there exists a true distribution | in time index , and a probability distribution generated by the generator . The goal of this minimal-maximization game is to | minimize the generation error of the generator, while maximizing the discriminators ability to distinguish false ratings from true ratings. Unsupervised learning can also be applied to classify different types of learning styles to recommend the most suitLiang et al. presented a MEUR model based on Graph Convolution Network and reinforcement learning. The author considered the learning process as a MDP. Applying a user-centric reasoning method and maximizing cumulative reward by actor-critic algorithm, is defined as below: () = 1
id: 2abffd8f3519b3b6ab209c55eacd886e - page: 11
=1 ,=0 (|, ()) (+1), where is the parameter of the Actor network, is the number of samples, is the sample index, and are the time step and action indexes, (|, ()) is the probability of , (+1) is the immediate reselecting action in state ward obtained in state +1 , and is the tuning parameter. (13) Y. Lin et al.: Preprint submitted to Elsevier Page 11 of 19
id: 6d375693d7b3e2118f1baac1dc1678a8 - page: 11