Research On Personalized Education Service Online Based On Deep Learning

Education is the grand plan of a country.The General Plan for Deepening Educational Evaluation Reform in the New Era clearly states that higher education institutions should innovate evaluation tools and use modern information technology such as artificial intelligence and big data to explore the whole process and elements of student learning evaluation;and improve the use of evaluation results to play a comprehensive role in guidance,identification,diagnosis,regulation and improvement.In recent years,due to the COVID-19,major educational institutions and universities have been fighting to accelerate the construction of online course resources and platforms at the request of the National Headquarters.Problems such as how to provide,access,and improve personalized online education services need to be solved,which makes the key technologies behind personalized online education services,such as artificial intelligence and big data,a research hotspot in academia.Personalized online education services are embedded in all aspects of "learning,practice,examination".As a learner,when faced with a huge amount of learning resources,I am eager to know whether this course is what I am looking for.In the process of learning,I wonder whether brushing these questions can improve my level,and whether I can pass the final exam successfully.In view of the above challenges,this paper aims at the most critical data of students’ learning behavior in online education,and tries to provide updated technical solutions for online personalized education services based on the latest theories and technologies related to deep learning.To achieve more accurate and interpretable prediction of students’ scores,personalized recommendation of learning resources and knowledge state modeling.The main research contents of this paper are as follows:1.Course recommendation based on deep learningCourse recommendation is to use data and algorithms to intelligently recommend courses or learning materials for users to help them find what they need quickly.On the basis of a comprehensive overview of deep learning-based course recommendation methods:(1)This paper proposes an auto-encoder-based course recommendation model.Existing course recommendation models do not distinguish the importance of various learning behaviors well,ignore the influence of personal preferences,and to certain extent have the problem of cold start,so,a course recommendation model based on autoencoder is proposed.The model first mines the potential learning preferences of learners through an autoencoder with an attention mechanism.Then,the course relevance decoder is used to construct the relevance information of the course.Finally,the course recommendation results are obtained by combining the learner’s learning preference information and the course association information.Experiments show that the proposed model is more accurate than the benchmark course recommendation model.(2)This paper proposes a course recommendation model based on a multilayer attention-gated recurrent network.The existing session-based course recommendation model ignores the fact that the courses in each session of a user are actually highly homogeneous,and highly heterogeneous between different sessions.At the same time,learners’ preferences actually change over time rather than being static,so a course recommendation model based on a multi-layer attention-gated recurrent network is proposed.The model slices user behavior sequences into multiple historical sessions as input,and then uses a triple-formed layer attention mechanism to extract users’ interests in each session,users’ dynamic preferences,and users’ short-term preferences,respectively,to finally obtain a mixed preference representation of users.Experiments show that this method outperforms other mainstream advanced methods.2.Knowledge tracing based on deep learningThe main task of knowledge tracking is to extract potential learning patterns from learners’ historical learning trajectory information and build a model of their knowledge state over time,so as to judge learners’ knowledge mastery and predict their future answer performance.On the basis of a comprehensive overview of deep learning-based knowledge tracking methods:(1)This paper designs a knowledge tracking method based on convolutional neural networks and recurrent neural networks.To address the weakness of convolutional neural networks,which are better in modeling local features but weaker in long-term modeling,they are combined with recurrent neural networks with long-term modeling capability and applied to knowledge tracking to model both local and long-term features.Extensive experiments on publicly available datasets show that the model combined with recurrent neural networks works better,but with higher complexity.Subsequently,combining convolutional neural networks with attention mechanisms was tried and found to be less effective than combining with recurrent neural networks,but with lower algorithmic complexity.(2)This paper proposes a new knowledge tracking model based on a self-attention mechanism.To address the problems of long-term dependency,poor interpretability,and lack of learning features in the field of deep knowledge tracking,a new knowledge tracking model based on the Transformer model with a self-attention mechanism is proposed.It models the forgetting behavior during learning by using time interval information instead of the relative position encoding used by the original Transformer,and enriches the learning feature information by adding topic features to the input.The experiments show that the model has a large improvement over the benchmark model.3.Student performance prediction based on deep learning(1)This paper proposes a performance prediction model based on multi-feature fusion.Existing performance prediction models either use traditional machine learning methods to extract simple low-order features or use deep neural networks to extract high-order features while losing some information in low-order features when extracting learned features,so,a performance prediction model based on multi-feature fusion is proposed.The model obtains both learned low-order features as well as high-order features by combining different deep network structures.Experiments show that the model has higher prediction accuracy.(2)This paper proposes an achievement prediction model based on a self-attention mechanism.To address the shortcoming that recurrent neural networks can only extract the temporal information of learning behaviors but not the overall information in performance prediction,a performance prediction model based on the self-attention mechanism is proposed.The model mainly includes a temporal behavior feature generator based on gated unit neural network,an overall behavior feature generator based on deep neural network,and a feature fusion mechanism with attention mechanism.The experiments show that the model can more realistically reflect the learning status of the learners and achieve more accurate performance prediction.