Course-oriented Expertise State Model And Application

Teaching evaluation is an important part of the learning process and directly reflects the individual differences of students.However,the GPA-style summative evaluation in colleges and universities cannot understand students’ knowledge mastery in different courses in detail,and it is time-consuming and labor-intensive to analyze students’ history courses one by one.To address this problem,this thesis firstly proposes a course classification method based on the distribution of students’ performance,which can reduce the number of courses to be analyzed for teaching evaluation while maximizing the differences among courses? secondly,the classified courses are used as a guide to build a professional knowledge state model so that teachers can quickly understand students’ knowledge level.Among them,the expertise state model based on course relevance accomplishes the task of predicting students’ future performance? the expertise state model based on monotonic attention mechanism accomplishes the task of tracking the change of students’ knowledge level,which can guide universities to personalize the training of students.The main work of this thesis can be summarized in the following three aspects.(1)Proposing a method for classifying courses based on the distribution of student performance,and using course categories as a guide for educational evaluation:The differences in difficulty,differentiation,and content among courses are reflected in the distribution of grades of the students taking the courses,so the degree of correlation between courses can be reflected by calculating the correlation coefficient between the distribution of grades corresponding to the courses.In this thesis,we propose a hierarchical clustering model based on the overlapping correlation coefficients of courses to classify courses,and combine the elbow index to determine the number of clusters for the course clustering task and complete the unsupervised classification task of courses.In order to cope with the lack of student performance in new courses,a decision tree model is used to explore the relationship between basic attributes of courses and course clustering categories,which solves the ”cold start”problem of new courses.(2)Propose an expertise state model based on course relevance to solve the task of predicting students’ future performance:Student performance is influenced by students’ knowledge status,course status,and class setting.In this thesis,we use the XGBoost model to predict the class average using course and class characteristics,and propose the concept of ”course class composite index” to control the confounding influence of course and class situation on students’ course performance? we design a self-coding network-based performance prediction module to input students’ expertise status and A self-coding network-based grade prediction module was designed to predict students’ future grades by inputting their expertise status and course class composite indicators.Experiments on real data sets are also conducted to verify that the proposed method has high accuracy in the prediction task.(3)Proposing an expertise state model based on monotonic attention mechanism to solve the problem of tracking students’ knowledge level changes in traditional classroom teaching scenarios in universities:Students’ knowledge level is a dynamic concept,which is reflected as students’ expertise state at different times.In this thesis,we construct an expertise state model consisting of a course encoder,a knowledge encoder,a knowledge retriever and a response prediction network? we design a monotonic attention mechanism to act on the encoder and knowledge retriever,simulate students’ ”forgetting behavior” and deal with the time-series modeling problem of historical course grades.The effectiveness of the model for tracking students’ knowledge level changes in different course categories is also experimentally verified on real data sets,and the model’s stable performance prediction ability in different knowledge domains is verified on different data sets.In summary,this thesis addresses the shortcomings of GPA-style summative evaluation in colleges and universities,and proposes a course classification method that can be easily extended to meet the needs of detailed analysis of students’ knowledge levels in different courses in educational evaluation? then,a knowledge state model based on course relevance and a knowledge state model based on monotonic attention mechanism are established with course categories as the guide,which accomplishes the two major tasks of predicting students’ future performance and tracking changes in students’ historical knowledge levels,and meets the needs of transferring learning evaluation from the past simple summative evaluation to a wisdom evaluation model from the perspective of knowledge state modeling.