Cognitive Diagnosis And Reinforcement Learning Based Recommendation Strategy For Learning Material In Dual Objective Adaptive Learning

The Ten-Year Development Plan for Education Informatization(2011-2020)proposes to respect the individual differences of each learner and use current science and technology to provide learners with appropriate education.In order to maximize the comprehensive and personalized development of students,it is necessary to provide personalized learning assistance adapted to the various characteristics of students,and the emergence of adaptive learning provides a way to achieve this personalized learning.For adaptive learning to work effectively,the core is to build a suitable learning material recommendation strategy.The construction of a good recommendation strategy is inseparable from the support of cognitive diagnosis,which provides a method for measuring learner characteristics for adaptive learning,and reinforcement learning,which provides a technology to improve the recommendation efficiency for adaptive learning.At present,the research on adaptive learning material recommendation strategy based on cognitive diagnosis and reinforcement learning mainly starts from the learner’s attribute mastery,and does not consider the learner’s ability factor.Considering learners’ ability factors is more conducive to recommending personalized learning materials for learners(Ren Weiwu et al.,2020).Therefore,this study considers the combination of learners’ attribute mastery and ability to apply to the research of adaptive learning material recommendation strategy to improve the recommendation effect and recommend more suitable learning materials for learners.Based on Tang(2019)’s Q-learning recommendation strategy(TQ),this study combines learners’ attribute mastery and ability factors,and develops a dual-objective Q-learning recommendation strategy(DQ-AA)based on learners’ attribute mastery and ability factors.For DQ-AA recommendation strategies,this study first verifies the effectiveness of DQ-AA by comparing DQ-AA with random recommendation strategies and TQ.Secondly,the influence of three factors,question length,exploration and utilization parameters,and learning times,on the recommendation effect of DQ-AA was explored.Finally,the application effect of DQ-AA in complex learning situations is explored.The main conclusions of this study are as follows:(1)DQ-AA has an ideal recommendation effect in the adaptive learning system,which can not only provide learners with learning materials that conform to their attribute mastery,but also recommend learning materials of appropriate difficulty for learners considering their ability factors.(2)The performance of the DQ-AA recommendation strategy in this study was much better than that of the random recommendation strategy,and the DQ-AA recommendation strategy could recommend more appropriate learning materials for learners compared with the TQ recommendation strategy.(3)DQ-AA will be affected by the length of the questions and the number of learnings,among which the number of questions and the number of learning times are large,and the specific and better recommendation effect of DQ-AA.(4)In complex learning situations,DQ-AA still has a better performance,and DQ-AA is more affected by exploration and utilization parameters.This shows that the performance of DQ-AA in practical application scenarios is still relatively stable.