Research On Knowledge Tracing Method Based On Attention Mechanism

The traditional teaching methods and model teaching have become the bottleneck restricting the healthy development of teaching.In accordance with the spirit of "China’s Education Modernization 2035",the development and application of big data,artificial intelligence,the Internet of Things,and other information technologies have created conditions for changing the traditional teaching model,improving the intelligence level of the teaching system,and made it possible for students to discover blind spots in learning.To accurately discover blind spots in teaching and learning,the knowledge tracing tasks of studying students’ learning status and predicting future performance have become a research hotspot.The existing works show that considering concepts from exercises when researching knowledge tracing tasks will help students allocate energy and time more efficiently.The attention mechanism has been proved to be very effective in analyzing and predicting student performance.Researchers regard the knowledge tracing tasks as being influenced by three factors: exercise knowledge,problem knowledge,and knowledge mastery.Meanwhile,they take the time distribution of the exercises into account,which improves the accuracy of the modeling of existing students’ performance.At present,there are few studies analyzing the dependencies between concepts and contextual knowledge mastery.The existing studies still have the following deficiencies.(1)The knowledge tracing methods only consider the influence of a single knowledge concept when analyzing the students’ problem-making status,which results in information transfer bias in the simulated learning process,and affects the accuracy of predicting students’ future performance on the problem;(2)The knowledge tracing methods are used in the simulation.When the students’ knowledge mastery state is neglected,the behavior of memory and partial forgetting is difficult to accurately assess the students’ knowledge mastery state in the context.The work in this paper aims to improve the knowledge tracing method based on attention mechanism from two aspects of "diversified distribution" and "context information",according to the deficiencies of knowledge concept distribution and modeling knowledge mastery in existing research.The main research works in this paper are as follows.(1)Knowledge tracing method based on diversified conceptual attentionThe existing methods only consider the impact of a single concept on emerging problems,while ignoring the interaction among a large number of concepts in the exercises.Therefore,this paper constructs a network called the Diversified Concept Network(DCN),which marks the interrelationship of various knowledge concepts in the exercise,and combines the Diversified Semantic Network(DSN)to improve the performance of existing knowledge tracing methods.Firstly,the answer data is obtained from the intelligent tutoring systems,and the dependence of the textual information of the exercises is analyzed.Secondly,the knowledge elements contained in students’ answers are extracted through DCN.Thirdly,DSN is used to extract the similarity of a large number of knowledge elements and integrate them into knowledge concepts.Then,calculate the attention score between the knowledge concepts required by the question to be tested and the knowledge concepts included in the exercises that the students have answered.At the same time,it predicts students’ response performance to emerging questions to formulate reasonable learning strategies.Finally,compared with state-of-the-art knowledge tracing methods,experiments on four public datasets validate the effectiveness of our method.The results show that optimizing the attentional mechanism by considering rich knowledge concepts can improve the accuracy of the knowledge tracing model in predicting student performance.(2)Knowledge tracing method based on bidirectional memory-enhanced attentionThe existing methods only consider the unidirectional transmission of knowledge acquisition,but ignore the characteristics of students’ memory and forgetting.Therefore,this paper constructs a network called the Bidirectional Memory-Enhanced Network(BMEN),which focuses on analyzing the interaction between the knowledge mastery state before and after each learning moment,and performs a bidirectional memory enhancement of the knowledge mastery state to improve the existing knowledge tracing method.performance.Firstly,the answer data is obtained from the intelligent tutoring systems,and the knowledge concepts contained in the exercises are analyzed.Secondly,the past and future knowledge mastery states are obtained through BMEN,and the knowledge mastery state after the influence of context information is updated.Then,calculate the attention score between the knowledge concepts required by the question to be tested and the knowledge concepts included in the exercises that the students have answered.At the same time,the model predicts students’ performance on emerging exercises to develop reasonable learning strategies.Finally,compared with state-of-the-art knowledge tracing methods,experiments on four public datasets validate the effectiveness of our method.The results show that knowledge acquisition by considering memory and forgetting can improve the accuracy of the knowledge tracing model in predicting students’ performance.