Research On Data Driven Personalized Learning Optimization

“Learning” is a complex and cumbersome event.Individual’s learning effect is affected by many factors.Moreover,different individuals have different learning habits.Especially in the emerging mode of network-based learning,students face many challenges,such as the complexity of learning resources,the liberalization of learning time,the diversification of learning status and so on.Most students lack awareness of their own learning characteristics and reasonable planning of time and learning tasks.In the face of a large number of learning tasks and fragmented time,it is usually difficult to make effective time or learning task arrangements.Unreasonable time arrangement may reduce the efficiency and quality of task completion,and eventually lead to limited learning effect,increased learning pressure,and even affect students’ mental health.Therefore,there is an urgent need to study suitable personalized learning optimization schemes according to students’ learning characteristics,so as to help students optimize learning efficiency and harvest the best learning effect.Although there has been a lot of research work on learning optimization,there are still three main challenges:(1)The existing work has not yet proved the influencing factors of learning.On the one hand,the data adopts the way of questionnaire survey,on the other hand,it uses the means of case analysis,which is subjective and one-sided.(2)Based on online learning data,there is a lack of research on the quantification of learning efficiency,as well as the influencing factors of learning efficiency and their impacts.(3)Although the existing research on learning optimization has put forward many universal theoretical optimization strategies,it ignores the differences between individuals and cannot provide a calculation method to form a specific learning task schedule.In view of the above challenges,this paper studies personalized learning optimization based on students’ learning behavior data.Specifically,this paper does the following parts:1.More comprehensive characteristics of learning influencing factors are extracted.Two real online learning data sets are used.Among them,Mooccubex dataset extracts 21 influencing factor features from four aspects: learners’ own ability,learning attitude,learning preference and learning behavior;The Student Data dataset extracts the characteristics of 13 influencing factors from three aspects: student users,learning tasks and time slices.Through data analysis,the influence of these characteristics on learning effect is verified in detail.2.A prediction model of learning efficiency based on extremely randomized trees is proposed.Firstly,the characteristics of the influencing factors are analyzed in detail,and the correlation between these factors and learning effect and learning efficiency is calculated;Then,the top 12 important factors are selected as the input of the model.By comparing six classical machine learning algorithms,the most suitable extremely randomized trees algorithm training model is selected,so as to predict the learning efficiency of learners in various learning tasks in a certain period of time;Ablation experiments were carried out to verify the importance of the extracted influencing factor features on learning efficiency.3.A personalized learning task allocation model based on bipartite graph is proposed.The bipartite graph method is used to construct the learning task allocation scenario,and the adaptive utility function is designed according to different learning objectives.Based on this,a dynamic allocation algorithm TLTA based on transfer learning is proposed to formulate a reasonable task allocation scheme for students;Finally,a large number of experiments are carried out on real student data sets to verify the effectiveness and applicability of the scheme.