Research On Knowledge Tracing Method Based On Deep Learning

With the continuous development of the Internet and computer technology,online education has become an important branch of the education field.Online education provides students with wider,more flexible,and more convenient education,especially during the period of the global spread of the COVID-19 pandemic,its advantages over traditional teaching methods are more significant.Automatic assessment of students’ mastery of knowledge in online education is an important task.Knowledge tracing is the task of modeling students’ mastery of knowledge based on their past answer history,so that we can accurately trace their mastery of the knowledge points.As a basic task of intelligent education,knowledge tracing plays a crucial role.Not only can it effectively trace students’ mastery of knowledge,but it has also been widely used in student learning status detection,personalized test recommendation,personalized teaching and other fields.This has high practical significance.With the flourishing development of deep neural networks and online education,knowledge tracing research has received a lot of technical and data support and has attracted a lot of attention from researchers,achieving good research progress.However,in the existing research,most knowledge tracing methods have not considered the fact that students with similar answer experiences have comparability in their mastery of knowledge,and have not made full use of the collaborative information of similar students to better model students’ knowledge states.At the same time,we found that in the data collection phase of research,there is an unavoidable "answer bias",that is,there is a serious imbalance in the number of correct and incorrect answers related to a certain concept practice.Existing models excessively rely on this bias information to improve model performance,but this is not the true goal of knowledge tracing.In response to the above problems,this paper focus on the method research of knowledge tracing around how to make full use of collaborative information and solve the problem of excessive reliance on "answer bias" information.In real scenarios,when a teacher evaluates whether a student can answer a question correctly,they not only consider the student’s historical learning status,but also take into account the learning status of similar students.To address how to obtain and efficiently utilize collaborative information,this paper proposes a collaborative knowledge tracing method based on self-supervised learning.This method designs a method based on the overlap rate of answers from similar students to retrieve similar students,and uses an attention mechanism to obtain the common knowledge status of similar students.By fusing the common knowledge status of the target student and similar students using a gating mechanism,the final knowledge status representation is obtained and used for prediction.In addition,to ensure the consistency between the common knowledge status and the target student’s knowledge status,this paper introduces self-supervised learning and treats them as a positive pair to obtain a better knowledge status representation.The evaluation results on three commonly used datasets show that this method has significant performance improvement compared to previous methods.In response to the problem of excessive reliance on "answer bias" information in existing methods,this paper proposes a knowledge tracing method based on counterfactual reasoning.It analyzes the knowledge tracing problem in the context of causal relationship,establishes the causal graph of the problem,calculates the total causal effect and the direct causal effect of "answer bias" information in the training phase.Finally,the direct causal effect of "answer bias" is subtracted from the total causal effect,effectively alleviating the model’s over-reliance on "answer bias".