The Research On OJ Programming Exercises Recommendation Algorithm Based On Knowledge Graph

Online Judge(OJ)system provides a platform for programming users to practice independently,while a large number of programming exercises in the OJ system makes it difficult for users to choose the right one.Therefore,OJ programming exercises recommendation plays an important role for programming users,which can assist them in the selection of exercises to save time for them.In this paper,first of all,according to the passing rate of students to do the programming exercises recommendation,the students with high passing rate are easy to do right,which can increase their interest in doing the exercises.So the purpose of the algorithm is to recommend programming exercises with a high passing rate to students.The algorithm uses the Deep Knowledge Tracing(DKT)model to predict the pass rate of the exercises.The original input of DKT is changed to the exercises number,additional information is added,and the input is compressed.The experimental results show that the prediction effect is better than that of the DKT model.Then,recommendations are made according to the prediction results to improve the interpretability of the recommendation algorithm.Secondly,on the basis of traditional recommendation algorithm,this paper combines collaborative filtering with content-based recommendation for hybrid recommendation.By introducing the programming exercises text information into collaborative filtering,the defect of ignoring the content information of the topic itself is made up.OJ programming exercises have the characteristics of corresponding text description information and several knowledge points.In this paper,OJ programming exercises are regarded as a small knowledge map,and the text is vectorized by using Bert(bidirectional encoder representations from transformers)model,which is integrated into collaborative filtering to improve the recommendation results.The specific research contents of this paper are as follows:(1)First,we predict the passing rate of students.In this paper,DKT model is used to predict the passing rate of exercises.The existing DKT model regards the programming exercises under the same knowledge point as the same problems,which makes it impossible to predict accurately according to the characteristics of each exercise.Therefore,the selection of the subject input,at this time the output is the probability of each subject through,help to improve the prediction result.But this method will also lead to the problem that the input exercises are independent of each other and the input dimension is too large.In order to solve these two problems,the algorithm in this paper first increases the knowledge point information in the input,establishes the relationship between the exercises,and then compresses the input dimension to prevent the excessive dimension from affecting the efficiency of training.In the test of Codeforces(CF)website data set,the prediction result of the improved model is improved.(2)The knowledge graph of OJ programming exercises is constructed.The knowledge point,programming exercises and difficulty information provided by CF website are used as the entity of the knowledge graph,and the relationship between them is defined manually.The relationship between the knowledge points is defined according to the expert opinions and algorithm competition textbook.(3)The text description of each exercise and some knowledge points contained in it are regarded as a small knowledge graph,and the data information is used to embed the exercise text.The whole process is regarded as text classification.The text vector of the exercise is obtained by using the best model and the subsequent fine-tuning layer,and then the vector is used in the collaborative filtering algorithm for hybrid recommendation.(4)Based on the above algorithm and work,this paper proposes two strategies of programming exercises recommendation.One is to use the predicted pass rate to recommend the exercises with high pass rate according to the ranking results.The second is to use knowledge graph for exercises recommendation,using the concept of knowledge graph to vectorize the topic text,and integrating the text content information into the collaborative filtering algorithm for hybrid recommendation.