Research On Deep Learning-based Knowledge Tracing Methods

Knowledge Tracing(KT)is an essential task in educational data mining field,the content of knowledge tracing is to model a student’s mastery of knowledge by his/her past problem solving sequences,so as to predict student’s learning performance of an essential task of the system is to understand student’s mastery of knowledge.In Intelligent Tutoring Systems(ITS),one of the important tasks is to understand students’ mastery level of concepts,only fully understanding every student’s situation can the system operates targeted,personalized tutoring.Intelligent Tutoring System obtains student’s present performance through knowledge tracing,then the system can improve student’s learning performance by those functions like personalized exercise recommendation.Thus,studying knowledge tracing task is of great significance.Since knowledge tracing task is proposed,it has attracted a variety of attention in both the academia and the industry,plenty of knowledge tracing methods are introduced including classical Bayesian methods and deep learning-based methods.It should be noted that since deep learning-based methods are introduced,they have surpassed traditional Bayesian methods in a large margin,which have become the main interest of the academia.However,most of existing deep learning-based methods still face the problem of relatively weak modeling capability,for example,convolutional neural network-based knowledge tracing methods excels at modeling local features,but they need to improve when dealing with long exercise sequence modeling;recurrent neural network-based methods perform outstanding long exercise sequence modeling capability,yet their relatively weak local features modeling problem is still left to solve.Aiming at problems above,this thesis proposes two deep learning-based knowledge tracing methods:(1)Considering the local feature modeling capability of convolutional neural network and the ability of long term sequence modeling of recurrent neural network,the thesis proposes a convolutional neural network and recurrent neural network-based knowledge tracing method,which combines convolutional neural network with recurrent neural network.The model treats student’s exercise sequence as raw data,then preprocesses the data as the input of the model.The convolutional neural network first models the preprocessed data,and the input of the recurrent neural network is the output of the convolutional neural network,the recurrent neural network performs long term sequence modeling,the output of the recurrent neural network will be operated with linear dimensional reduction.Extensive experiments on two public datasets of the model shows improvement of the model over the other state-of-the-art knowledge tracing methods in three classic indicators,which proves the effectiveness of the model.(2)It is well-known that the attention mechanism also has excellent long sequence modeling capability,this thesis proposes the attention augmented convolutional knowledge tracing method,which appends the attention mechanism after the output of the convolutional neural network in order to effectively choose the results of the output,and the output of the attention mechanism will be treated as the prediction of the model by linear dimensional reduction operation.Extensive experiments on public datasets show that the mechanism augments the predicting ability of the model,moreover,the training time of the model is less than other knowledge tracing methods.Compared to other state-of-the-art deep learning-based knowledge tracing methods,the method is better among those methods,which validates the feasibility of the method.