基于序列遗忘行为的动态键值记忆网络

Dynamic key-value memory network based on sequence forgetting behavior

知识追踪过程中动态键值记忆网络(dynamic key-value memory networks,DKVMN)未考虑答题序列对学生学习和遗忘行为的影响,导致DKVMN预测性能仍有一定的提升空间.针对该问题,提出一种基于序列中遗忘行为的动态键值记忆网络(dynamic key-value memory networks-forget,DKVMN-F).将重复时间间隔视为遗忘因素,过去答题次数视为学习因素,利用门控机制在DKVMN的预测层整合遗忘因素和学习因素.仿真结果表明,相较于同类模型,DKVMN-F模型的曲线下面积(area under curve,AUC)和准确率(accuracy,ACC)分数更高,更加适用于知识追踪任务.

In the process of knowledge tracking,dynamic key-value memory networks(DKVMN)do not consider the influence of the answer sequence on students'learning and forgetting behavior,which results in some room for improvement in the prediction performance of DKVMN.A dynamic key-value memory network(dynamic key-value memory networks-forget,DKVMN-F)based on the forgetting behavior in the sequence is proposed.The repetition time interval is regarded as the forgetting factor,and the number of past answers is regarded as the learning factor,and the forgetting factor and the learning factor are integrated in the prediction layer of DKVMN by using the gating mechanism.The simulation results show that,compared with similar models,the DKVMN-F model has higher AUC(area under curve)and ACC(accuracy)scores and is more suitable for knowledge tracking tasks.