面向在线教育考试评估的数据挖掘技术研究与应用

With the rapid development of information technology in the domain of education, a massive number of online examination data are being recorded, from which we could design the data-driven models for evaluating examination and assisting teaching. Unfortunately, it is hard for existing models to precisely capture the characteristics of testing problems and the proficiency of students from these examination data that are sparse, high-dimensional and noisy. Meanwhile, the personalized requirements of students and teachers are usually ignored in these studies. To this end, with the help of the techniques like recommender systems and the domain knowledge of education, this project studies the key issues on the data mining solutions for examination evaluation of online education, and these issues are related to testing problems, students and teachers, respectively. In this way, the skill proficiency of students could be improved. Specifically, we first develop a model, which combines the fuzzy theory, to analysis and evaluate the characteristics of both subjective and objective testing problems. We then solve the profiling problem and the personalized studying materials recommendation problem for students, and our solutions consider both the cognitive diagnosis and the context of the students. On this basis, we further study the way of personalized teaching assistance for teachers, and meanwhile, we propose algorithms to help automatically form collaborative learning teams in a class. Finally, the above discoveries will be applied to online education system "Zhixue.com". In summary, this project will propose effective solutions for the data mining problems in examination evaluation. As a practice of " Internet+Education", not only the capability of educational data mining for online education could be enhanced but also the application areas of recommender systems could be expanded. Thus, the research of this project has important theoretical and practical value.

随着教育信息化进程的深入，学生的在线考试数据得到不断积累，为数据驱动的教育考试评估和辅助教学提供了良好条件。然而，已有模型很难从海量、稀疏、高噪的数据中对试题特征和学生学业水平进行准确分析，且较少考虑学生及教师的个性化需求。针对以上问题，本项目以教育考试中涉及的试题、学生、教师为对象，以个性化推荐等技术同教育领域知识相结合为手段，以提高学生学业水平为目标，开展面向在线教育考试评估的数据挖掘技术研究。具体地，结合模糊集理论，构建主客观题联合评估模型；基于认知诊断和情境感知的推荐算法，实现学生画像和个性化学习资源推荐；设计针对教师的教学建议和指导，解决面向协同学习的学生自动分组问题。并将研究成果应用于在线教育系统“智学网”。本项目提出了面向教育考试评估的数据挖掘有效解决方案，提高了数据挖掘对在线教育的辅助能力，拓展了推荐系统的应用领域，是“互联网+教育”的具体实践，具有重要理论意义和应用价值。

随着教育信息化进程的深入，对智能教育提出了更高的要求。通过积累学生的在线考试数据，研究数据驱动的教育考试评估和辅助教学方法具有十分重要的意义。然而，已有模型很难从海量、稀疏、高噪的数据中对试题特征和学生学业水平进行准确分析，且较少从辅助教学的角度考虑学生及教师的个性化需求。针对上述问题，本项目以教育考试中涉及的试题、学生、教师为对象，以联合评估模型和认知诊断方法为手段，开展面向在线教育考试评估的数据挖掘技术与应用研究。首先，研究教育试题的评估分析方法，揭示试题之间与技能之间的关联关系，从试题深度表征、相似试题发现和试题难度预测三个方面进行展开，提高试题评估的有效性。其次，设计量化学生认知状态的建模方法，追踪学生长期学习状态，实现学生成绩趋势的预测。在此基础上，根据对试题和学生的研究，提出个性化试题推荐等方法提高了教学辅助系统质量。同时，研究基础方法以便于后续在线教育在试题质量评估、学习资源推荐等场景的应用实现。在本项目的支持下，已发表学术论文52篇（IEEE/ACM Trans系列12篇，CCF推荐的A/B类会议35篇），申请发明专利6项，代表成果获数据挖掘领域最好的国际会议KDD 2018的年度最佳学生论文奖，项目负责人入选了中国科协“青年人才托举工程”（2017-2019）、获得第二届阿里巴巴达摩院青橙奖（年度全国十人）、教育部自然科学一等奖（2018）并于2019年获批基金委优秀青年科学基金项目。相关研究成果已在科大讯飞公司的在线教育系统“智学网”进行应用，相关成果还获得2020年度吴文俊人工智能科技进步奖一等奖。本项目通过探索在线教育的痛点，提出了在试题评估、学生认知诊断等方面的有效解决方案，有利于提高教辅系统质量，提升师生教和学的效率。