Research On The Method Of Knowledge Tracing For Blended Learning

With the continuous development of Internet technology,more and more people choose to acquire knowledge and skills through online education.Compared with traditional face-to-face teaching,online education has gained widespread attention because of its flexibility in time and place and the variety of contents.However,there are also some problems with online education,such as not being conducive to developing students’ communication and collaboration skills.Therefore,blended learning has emerged,which combines offline and online learning to achieve the integration and complementarity between online and traditional education and improve students’ learning experience and learning efficiency.When studying offline,students can choose learning resources that are suitable for their learning pace and level with the guidance and assistance of their teachers.However,when learning online,it is difficult for students to select suitable content from the vast amount of learning resources on online education platforms due to the lack of teachers’ assistance.To help students access learning resources suitable for them,online education platforms generally automatically assess students’ learning ability and knowledge level through methods such as data mining,and then recommend learning resources suitable for them based on the assessment results.Among them,Knowledge Tracing(KT),which can assess students’ knowledge level based on their historical answer records,is one of the common methods used by online education platforms.Although knowledge tracing is now widely used in various online education platforms,the existing knowledge tracing methods only focus on students’ online learning actions,ignoring the fact that in the blended learning scenario,students’ offline learning actions also lead to changes in knowledge states.Therefore,in the blended learning scenario,there may be large deviations between the knowledge state of students after going offline estimated by the existing knowledge tracing methods and the real knowledge state,resulting in the learning resources recommended by the online education platform for students after students have been offline for a period of time are not suitable for the real knowledge level of students.Moreover,online education platforms usually do not have access to students’ offline learning actions,so it is a challenge to evaluate students’knowledge states after offline without offline learning data.To address this challenge,this dissertation proposes a new knowledge tracing model,the offline-aware knowledge tracing(OKT),which can mine the offline information from students’ online learning data recorded by online education platforms to estimate students’ knowledge states after offline.In addition,OKT can mine students’offline action preferences and make personalized predictions of students’ offline knowledge changes.Specifically,OKT first estimates students’ online knowledge states from online records,then estimates students’ offline trait based on their online knowledge states before and after offline,and finally uses offline trait to estimate students’ knowledge changes when they are offline,so as to update students’ knowledge states after offline.Experiments prove that OKT can mine the offline knowledge information in the online learning sequence.Moreover,the model performance of OKT outperforms existing knowledge tracing models.However,by analyzing the experimental results,it is found that OKT does not handle the offline cold start problem well.This is because there is less offline data during offline cold start,and it is difficult for OKT to mine students’ offline action preferences.For this reason,this paper further proposes OSKT,an offline knowledge tracing model based on knowledge structure modeling.first,OSKT updates students’ knowledge state on this knowledge concept through online records,and updates the knowledge state of other related knowledge concepts according to the dependency of knowledge structure.Then,OSKT estimates the offline trait of the corresponding knowledge concepts based on the online knowledge states of students before and after offline.Finally,OSKT updates the knowledge states of all knowledge concepts after offline based on the knowledge dependency relationships and the offline features of the corresponding knowledge concepts.It is found that OSKT is able to mine the offline information in the online learning sequence more effectively based on the knowledge dependencies,so OSKT achieves better prediction performance than OKT and is more effective in dealing with the offline cold start problem.Therefore,OSKT has higher application value in the blended learning scenario.