Research On Learner Clustering Method For MOOC Platform

In Massive Open Online Courses(MOOCs),learning style recognition is important to enhance the learning and teaching experience.In learning style recognition tasks,both clustering methods and hyperparameter optimization techniques play a very critical role.Clustering can bring similar samples together to better identify different learning styles.However,due to the large number of samples in MOOC datasets,the time cost of manually adjusting hyperparameters is very high,so it is necessary to use hyperparameter optimization technology to find the best combination of hyperparameters to improve the accuracy and generalization ability of the classifier.In this paper,two algorithms suitable for learning style recognition are proposed from the perspectives of optimizing common defects in clustering algorithms and combining learning style similarity measures,namely,the improved clustering algorithm for learning style recognition in the MOOC environment and the adaptive weight particle swarm optimization algorithm based on learning style.(1)An improved clustering algorithm for learning style recognition in MOOC environment is proposed.First,the number of clusters is determined by the elbow method,and two threshold definitions based on distance measures for the canopy algorithm are proposed.The fast partitioning feature of the canopy algorithm is then used to find the initial center point in each cluster,and clustering is performed using the K-mean algorithm.Learners’ different preferences for different learning styles are used to define learners’ global feature vectors to automatically identify learning styles.Finally,the performance of the proposed method on two datasets is verified using four classifiers.From the performance evaluation of the classifiers in the experimental results,it can be found that the improved clustering algorithm for learning style recognition in the MOOC environment proposed in this paper can It overcomes the problem of unstable clustering results caused by random selection of K values and initial cluster centers in the K-means clustering algorithm,and effectively improves the performance of classifiers in learning style recognition tasks.(2)An adaptive weight particle swarm optimization algorithm based on learning style is proposed.First,the learning style similarity score is proposed based on the cosine similarity matrix and used to initialize the sample weight.It is also used as the initial weight parameter of the adaptive weight particle swarm optimization algorithm.Then,in the iterative process of the particle swarm optimization algorithm,the prediction error value of the classifier on the training set is calculated to update the individual historical best hyperparameter combination and the global historical best hyperparameter combination.Finally,the above algorithm is applied in the process of learning style recognition.The experimental results show that the adaptive weight particle swarm optimization algorithm based on learning style measures the influence of the individual learning style of the sample on the performance of the classifier,and is applicable to all classifiers in learning style recognition,achieving stronger generalization ability and further Improves the performance of classifiers in learning style recognition tasks.