Researches Of Path Recommendation Algorithm For E-Learning

"Internet+education" is constantly changing the teaching mode of online education,personalized learning is considered an important learning method for future education.However,due to the coexistence of the homogenization of various learning resources and the insufficient supply of high-quality resources on the Internet,learners are faced with the problem of what kind of resources to choose in order to obtain the best results in the learning process.Learners' self-adjusting adaptive learning strategies to recommend appropriate learning resources based on learners' learning characteristics and preferences,and thus meet learners' personalized learning needs,these have become hot topics in the current online learning field.Based on a large-scale online learning environment,the thesis uses Long Short-Term Memory Network(LSTM)to construct a novel learning path recommendation method to provide navigation for students' personalized online learning to help them improve their learning efficiency.The main work mainly includes two aspects:A clustering-based learning path generation algorithm is proposed.This method calculates the similarity between learners based on the basic characteristics of the learners,divides the learners with higher similarity into the same cluster cluster,and then finds the learning of learners similar to the target user in the same cluster Path,according to the clustering result to generate the recommended learning path of the target user.This method can further provide input for LSTM training by making initial recommendations for learners and learning paths.An improved learning path recommendation based on Long Short-Term Memory Network is proposed.This method trains an extended long-short-term memory network model,recommends the nearest learning path to the learner and uses the model to predict the learning effect of the learner,and selects a suitable learning path to recommend to the learner according to the prediction result.Experimental results show that the proposed model has better adaptability and accuracy compared with other traditional recommendation models.