Research On OJ Problem Recommender Integrating Knowledge Status And Answer Willingness

In the online judge(OJ)system,users are able to solve programming problems that improve their programming abilities.Users may,however,find it difficult to find problems that are relevant to their programming abilities because they’re overwhelmed by the enormous amount of information in the OJ system.OJ problem recommender has been widely used in the OJ system because it allows users to mine their abilities and preferences and suggest the most practical problems for them.In contrast,the existing OJ problem recommender uses limited problem information,ignores user willingness,and considers only the user’s ability to solve the problem.This results in poor performance in the sparse problem interaction dataset.The purpose of this paper is to investigate the above problems and propose a recommendation system for OJ problems that integrates knowledge status and willingness to answer.Specifically,the research consists of:It is proposed to solve the shortcoming of ignoring problem information by incorporating knowledge status prediction models that incorporate problem information.Firstly,the statistical features of problems are analyzed.Text modelling uses the latent Dirichlet allocation to analyze the knowledge points of problems,and the comprehensive features of problems are formed from the statistics and knowledge points;Afterward,cyclic neural networks are constructed from the records submitted by users and the comprehensive characteristics of problems;Consequently,the entropy weight method is used to predict the user’s knowledge status more accurately by integrating the importance of problems into the output.An approach based on a graph convolutional network is proposed to address the problem of ignoring users’ willingness to answer questions.The first step is to build the answer willingness matrix from the user’s submission records and derive the initial features of each user node and problem node using the singular value decomposition process;At the same time,the user knowledge status and the comprehensive features of problems are used to calculate the similarity of user homogeneous nodes and problem homogeneous nodes respectively;A graph convolutional network is then used to propagate homogeneous and heterogeneous features,and the essential features are highlighted using self-attention;Finally,the feature propagation of homogeneous and heterogeneous nodes is carried out based on graph convolutional network,and the essential features are highlighted based on self-attention mechanism.The problem recommender strategy is designed that integrates the knowledge status and the willingness to answer questions.Give the user problems of appropriate difficulty according to his knowledge status to help him quickly improve his programming skills;According to users’ willingness to answer the problem,it is recommended that they responding to problems with high interest to maintain users’ enthusiasm for learning.Serial hybrid recommenders outperform single and parallel recommenders on a real dataset,according to an experiment.In a short period of time,it is able to improve the programming skills of a user quickly and steadily.