Research On Knowledge Tracing Method Based On Attention Mechanism

In recent years,with the rapid development of network and information technology,online education has gained rapid popularity.Today,the online education industry is transitioning from the digital and mobile era to intelligent education based on artificial intelligence and educational big data mining.The intelligent education system contains millions of students,and each student has different interests,potential characteristics,and learning ability.Smart education aims to provide students with more supportive learning guidance,analyze students’ weak knowledge concepts,and formulate self-adaptive learning programs for students,so as to realize personalized education.Therefore,how to model students’ learning process and accurately grasp their learning states using educational data mining and other related technologies is a very important topic.Knowledge tracing is the key technology to simulate students’ knowledge states,and it is one of the key research fields to enhance the ability of personalized education.As a cornerstone of student personalization modeling,knowledge tracing has achieved some good achievements in a long research accumulation,but it still faces many challenges due to the particularity of online education.The intelligent education system has a very large amount of resource data.How to find the decisive and important data in the massive data is a big problem.Online education allows us to learn on any device,anytime and anywhere,so students’ learning data is often sparse.Meanwhile,many knowledge tracing models do not design a reasonable model to use the text features of the exercises.Recently,attention mechanism has been applied to the field of neural network machine translation,and has achieved good performance.The main idea of the attention mechanism in deep learning is to simulate the human selective vision mechanism and select the information which is more critical to the current task from the massive information,which provides a new idea for knowledge tracing in smart education.Based on the attention mechanism,this paper improves the knowledge tracing model based on the long short-term memory network(LSTM)and dynamic key-value memory network.The main research work of this paper are as follows:(1)In the existing knowledge tracing model,the attribute information of questions is often ignored,which is just an implicit condition.In this paper,a heterogeneous information network is modeled for the attribute information contained in the questions,and a heterogeneous network embedding method based on meta-paths is designed to map the attribute relationship of the questions into the vector space through network representation,which effectively solves the problem of the sparsity of data.(2)According to the attribute characteristics of the question,combined with LSTM,a knowledge tracing model based on heterogeneous information network embedding and attention mechanism is proposed.By using the attention mechanism to capture the relationship between the students’ historical answer sequence and the current question,and by calculating the correlation degree between each question in the historical sequence and the current question,we can strengthen the most relevant questions in the historical sequence and improve the effect of knowledge tracing.In order to verify the effectiveness of the proposed model,we conducted experiments on multiple data sets,and the results show that the proposed model outperforms several comparative models.(3)We proposed a model named knowledge tracing with exercise-enhanced keyvalue memory network.Traditional knowledge tracing models use a specific number to represent the exercise,ignoring the rich semantic information contained in exercise texts.To solve the problem of difficult representation of learning resources,this paper puts forward a method of exercise representation based on semantic understanding.First of all,the BERT model is used to extract the vector representation of each word in the question text.Secondly,to increase the semantic connection between words,the word vectors are encoded and integrated by a bi-directional self-attention module to obtain the sentencelevel features of the whole question.Finally,we use a dynamic key-value memory network to simulate the learning process of students to help track their learning states.The experimental results on real datasets verify the effectiveness of the proposed method.To sum up,this paper studies the problem of knowledge tracing on the basis of attention mechanism and related algorithms of deep learning.At the same time,through a wide range of experiments,it is proved that our proposed algorithm can model students’ cognitive level more accurately,and then predict their future exercise-solving results.