引导式教学场景下深度强化学习的模型研究

Deep reinforcement learning model in heuristic coaching scenario

针对引导式场景,结合认知科学上学习区的概念,构造题库网络图,进而根据特定学习者的行为来划分割集,由此建立引导式教学场景下深度强化学习的模型,在推荐偏差指标的控制下,做出最适合学习者的内容推荐.对比实验证明了模型相比控制组能给出合理的"前向推荐",有效解决学习者作答正确率不稳定的问题.引导式教学场景下深度强化学习的模型能够拟合经验教师出题决策的思维方式,在历史作答数据中提取有效隐含信息,为学习者推荐最佳习题.模型亦可广泛应用在类似的引导式场景下.

By combining the concept of the learning zone in cognitive science,this paper constructs a network of question-base,where the cut set is made based on the behaviors of specific learners,for heuristic scenarios.A deep reinforcement learning model to make the best recommendation for learners is then proposed.The model is trained with the learner’s behavior under the control of the recommendation deviation factor,to export the best recommendation of content.A comparative experiment proves the model can effectively solve the unstable problem of the correct rate and outperforms the control group.The model imitates the thinking pattern of an experienced teacher,extracts valid implicit information in the historical answering data,and recommends the best exercises for the learner.The model can also be widely used in similar guidance scenarios.