Research On Online Judge Problem Recommendation Based On User Learning Behavior

With the continuous development of computer education,programming teaching has become a core course in computer courses.Many educational institutions use the Online Judge(OJ)system as an important academic tool for programming courses,which plays a crucial role in enhancing users’ programming skills,algorithm levels,and overall academic achievements.The OJ system usually maintains a large number of domain-specific problems(problem banks),however,these systems do not provide personalized recommendation services.The OJ system has rich evaluation data but lacks feedback on users’ learning outcomes,and the information overload of problems makes users lack direction in choosing problems.Our aim is to improve users’ learning efficiency by providing personalized learning suggestions and systematically organizing available resources.The specific work of this research is as follows:(1)To address the issue of how to measure the relationship between problems in the OJ system,we extract different users’ problem-solving records from three OJ systems in real-world scenarios.Based on this,we use social network analysis methods to extend our understanding of the correlation patterns and influence between problems and construct the interaction-related network of different problems in each OJ system.We quantitatively analyze the potential dependencies between different problems in each OJ system and provide problem-related recommendations to users based on the interaction network through modeling and analysis of each OJ system.(2)The OJ system usually lacks personalized problem recommendation services,which may result in users wasting time and energy when choosing problems.To solve this problem,we propose a time-window-based association recommendation mechanism by exploring the learning behavior of experienced learners at a fine-grained time scale.This mechanism adopts a collaborative filtering method that utilizes the analysis of problem associations within a time window and the improvement of time rule associations.By incorporating interaction information into the time window,problem-relatedness can be measured on the time dimension,resulting in better retrieval results.(3)To address the problem of missing problem knowledge s in the OJ system,we collect and integrate online problem-solving resources from authoritative OJ systems in China,and summarize multiple label information for each problem using the extracted label knowledge base.A problem recommendation method based on multi-label enhancement of the OJ system is proposed.Experimental results show that solving the problem of missing knowledge labels on OJ systems by using users’ learning records in the form of online problem-solving reports is more interpretable than solutions that solely rely on users’ behavior sequences.Compared with two different learning behavior sequences,the multi-label information of problems effectively improves the expression ability of sequence information,resulting in a significant improvement in recommendation performance.At the same time,in the behavior sub-sequences within a shorter time period,it has a significant advantage in recommendation accuracy and ranking quality.