一种知识图谱增强的在线评测系统习题推荐算法研究  

Research on Exercise Recommendation Algorithm for Online Judge System Enhanced by Knowledge Graph

作　　者：叶俊民[1] 宋家琦 张珂 于爽 YE Jun-min;SONG Jia-qi;ZHANG Ke;YU Shuang(School of Computer,Central China Normal University,Wuhan 430079,China)

机构地区：[1]华中师范大学计算机学院,武汉430079

出　　处：《小型微型计算机系统》2023年第11期2558-2565,共8页Journal of Chinese Computer Systems

基　　金：国家社会科学基金后期项目(20FTQB020)资助。

摘　　要：当前在线评测(OJ)系统中往往积累了大量的习题数据,学习者难以从海量习题数据中快捷、准确地识别出适合于自身的习题开展答题训练,只能大量地进行无差别刷题,习题与学习者的失配问题浪费了学习者大量的时间和精力,从而提升了学习者的学习成本、降低了学习效率.为此,本文提出了一种知识图谱增强的在线测评系统习题推荐算法,该方法结合了习题知识图谱,基于知识图增强推荐的多任务特征学习方法(MKR)模型完成习题推荐任务.首先,基于在线测评系统中的赛事集与习题知识体系构建习题知识图谱;其次,根据习题难度和学习者的能力水平,构建“学习者-习题”匹配矩阵;最后,利用交叉压缩单元,使用交替学习方法训练知识图谱嵌入(KGE)任务和习题推荐任务,完成学习者个性化习题推荐任务.在一个含有6919道习题、100名学习者的真实在线评测系统数据集上的实验表明,本文方法能够以84.2%的查准率完成在线评测系统学习者个性化习题推荐任务.At present,a large amount of exercise data is often accumulated in the online evaluation(OJ)system.It is difficult for learners to quickly and accurately identify the exercises suitable for themselves from the massive exercise data and complete the solution training.They can only brush a large number of undifferentiated questions.The mismatch between the questions and learners wastes a lot of time and energy,thus increasing the learning cost of learners Reduced learning efficiency.Therefore,this paper proposes an exercise recommendation algorithm of online evaluation system enhanced by knowledge graph.This method combines the exercise knowledge map and completes the exercise recommendation task based on the multi task feature learning method(MKR)model of knowledge map enhanced recommendation.Firstly,the exercise knowledge graph is constructed based on the event set and topic knowledge system in the online evaluation system;Secondly,according to the exercise difficulty and learners'ability level,the"learner exercise"matching matrix is constructed;Finally,the cross compression unit is used to train the knowledge graph embedding(KGE)task and exercise recommendation task by using the alternating learning method to complete the learner's personalized exercise recommendation task.The experiment on a real online evaluation system data set containing 6919 exercises and 100 learners shows that this method can complete the personalized exercise recommendation task of learners in the online evaluation system with precision of 84.2%.

关 键 词：习题推荐 学习分析 知识图谱 在线评测系统 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]