| In response to the outbreak of the COVID-19,many colleges and universities moved offline classrooms to online according to the policy of “suspend classes but not stop teaching and studying”,thus online learning has been more widely used.Adaptive learning plays an important role in many online learning systems.Adaptive learning is an educational method that uses computer algorithm to coordinate interactions with learners,and provide customized learning resources and learning activities to solve the unique needs of each learner.Compared with traditional classroom instruction,which usually provide all learners with the same learning materials,adaptive learning focuses more on individual differences.One of the main challenges it meets is how to provide learners with customized learning resources,that is,how to generate a personalized learning item recommendation list for learners in the massive learning resources.Knowledge tracking is often used to capture a learner’s knowledge level,and it can predict the learner’s performance on the next interactive learning item.Existing recommendations for learning items mainly depend on learners’ knowledge level captured by knowledge tracing,which reflects learners’ degree of mastery for learning items.Although these methods have made a great success in adaptive learning,there are still some challenges.For instance,（1）they fail to model the learner’s forgetting behavior and individualized differences in learning rate and comprehension ability when using knowledge tracking to obtain the knowledge level of learners;（2）For recommendation tasks,current methods considering the tasks of knowledge tracing and recommendation in separate steps,which makes the knowledge tracing is not contained in the final optimization goal,thus making it difficult for accurately capture learners’ dynamic knowledge level.Moreover,the recommendation only based on learner’s knowledge level will ignore learner’s preference for the type of learning resource,learning strategy and so on.To cope with the aforementioned issues,this paper explored the following work:Firstly,in knowledge tracking,to solve the forgetting phenomenon which occurs in the learning process and the learners’ individual differences,on the one hand,this paper designed a personalized forgetting controller,which uses three time-related features to strengthen the long and short-term memory network to better solve the learner’s complex forgetting problem.On the other hand,this paper used the learner’s personality to model individual differences,so that it could be take the learner’s learning and understanding ability into account when making knowledge level prediction.Therefore,the dynamic evolution of learners’ knowledge level could be more accurate.Secondly,to solve the separation of knowledge tracing and recommendation,this paper proposed a knowledge-enhanced multi-task learning course recommender framework to promote course recommendation.It regards knowledge tracking task and recommendation task as auxiliary task and main task,respectively.Through the information sharing mechanism between the two tasks,the knowledge tracking task can better help the course recommendation task.In addition,in order to not only focus on the learners’ knowledge level,but also consider the learners’ preference in the recommendation task,this paper adaptively fused the learners’ knowledge level,the learners’ sequence behavior and the personality to model the learners’ portrait.To ensure the logicality of knowledge structure when generating recommendations,this paper exploited a rule-based approach to select candidate courses.Therefore,the knowledge-enhanced multi-task learning course recom-mender framework could generate more personalized recommendation.Finally,to explore the applicability of the proposed model,this paper designed and developed a knowledge-enhanced multi-task learning course recommender system.The system has a friendly and concise human-computer interface.It can generate personalized course recommendation based on user’s knowledge level,personality and behavior,helping user to improve the learning efficiency.Besides,it can be used for collecting educational datasets containing personality to provide data support for subsequent research. |
