面向OJ训练的适应性学习路径推荐研究

The training of computational thinking is becoming more and more important. As a highly effective means to cultivate computational thinking ability, Online Judge (referred to as OJ) has not only been used in the programming contests, but also been applied to programming teaching and training. However, it is difficult for learners to choose the suitable problems when facing a great number of ones in OJ system. As a result, the training is inadequately effective. Hence, it is necessary to recommend the learning path to the learners. In this project, we aim at recommending the adaptive learning path for OJ trainees. The main research points are using knowledge graph and crowdsourcing to construct and supplement the knowledge model; using the code quality evaluation and learning behavior analyzing to construct the latent factor model which portrays the learner's professional competence; designing recommendation algorithms under the guidance of teachers. The result of this project will help increase the intelligence of online training and promote the deep integration of educational ideas and information technology.

计算思维能力培养的重要性日益凸显，作为一种卓有成效的计算思维能力培养手段，Online Judge（在线判题，简称OJ）不但用在了国内外程序设计竞赛中，更广泛用于编程教学和训练中。然而学习者面对OJ系统上海量的题目，很难从中选出符合自己当前水平的题目，达不到有效训练的效果，因此需要为学习者推荐学习路径。本项目研究面向OJ训练的适应性学习路径推荐，主要研究思路包括：采用知识图谱和众包机制构建和补全题库知识模型；结合程序代码质量评价和学习行为分析，构建刻画学习者专业能力的隐语义模型；设计教师参与指导的、能力训练主导的混合策略推荐算法。相信本项目的研究成果能更好地提升在线训练的智能性，帮助计算思维能力的有效培养，推进教育理念和信息技术的深度融合。

计算思维能力培养的重要性日益凸显，作为一种卓有成效的计算思维能力培养手段，在线判题（Online Judge，简称OJ）不但用在了国内外程序设计竞赛中，更广泛用于编程教学和训练中。然而学习者面对在线判题系统上海量的题目，很难从中选出符合自己当前水平的题目，达不到有效训练的效果，因此需要向学习者推荐适合的习题，帮助学习者快速提高程序设计能力。本项目以Online Judge习题推荐的设计为研究主线，围绕学习者建模、OJ习题建模和习题推荐策略等方面进行研究，提出了记忆与认知融合的个性化OJ习题推荐方法、基于深度强化学习与程序分析的OJ习题推荐模型和可用于程序分类与编程知识追踪的动态感知门控图注意神经网络等，取得较好效果，完成了预期的研究目标。项目产出学术论文8篇，培养1位博士研究生和2位硕士研究生。项目的成功实施，为更好实现习题推荐任务和培养程序设计人才带来新的启发，对其它学科和不同类型的习题推荐任务也有借鉴意义。