互助学习环境下可抗恶意评价的同伴互评算法  

Peer grading algorithm against malicious evaluation for collaborative learning

作　　者：赵鸣铭 王聪 李敏[1,3] Zhao Mingming;Wang Cong;Li Min(Computer Science College,Sichuan Normal University,Chengdu 610068,China;College of Movie&Media,Sichuan Normal University,Chengdu 610068,China;Network&Data Security Key Laboratory of Sichuan Province,University of Electronic Science&Technology of China,Chengdu 610054,China)

机构地区：[1]四川师范大学计算机科学学院,成都610068 [2]四川师范大学影视与传媒学院,成都610068 [3]电子科技大学网络与数据安全四川省重点实验室,成都610054

出　　处：《计算机应用研究》2020年第8期2305-2309,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61602331);四川省重点实验室开放课题(NDSMS201606);四川省教育厅重点项目(17ZA0322);四川省教育厅科研项目(17ZB0361)。

摘　　要：受学习者能力和意愿等主观因素的影响,互助学习环境下同伴互评结果与真实评价通常存在显著差距。为了提高同伴互评的质量,避免互评过程中利益驱动的恶意评价,引入少量由教师预评分的作业作为哨兵。通过评审人对哨兵的评审情况,以评审人的信誉评价作为权重向量,并利用阈值进行截尾,从而实现了对恶意评价的有效隔离。真实测评数据上的实验结果证明,相较于当前主流的互评算法,该算法能有效过滤恶意高评或低评,并且能适应较大数量级的学习者进行互评。在未来的研究中,将针对信誉模型的优化进行深入的研究。Influenced by subjective factors such as learner’s ability and willingness,there is usually a significant gap between peer review results and real evaluation in a collaboration learning environment.In order to improve the quality of peer review and avoid the interest-driven malicious evaluation in the process of mutual evaluation,this paper introduced a small number of teachers’ pre-scoring assignments as sentinels.Through the reviewer’s review of the sentinel,it used the reviewer’s reputation evaluation as the weight vector,and used the threshold for truncation,thus effectively separating the malicious evaluation.The experimental results on the real evaluation data prove that compared with other mainstream mutual evaluation algorithms,the algorithm can effectively filter malicious high evaluation or low evaluation,and can adapt to a larger number of learners to conduct mutual evaluation.In future research,an in-depth study will be conducted on the optimization of the reputation model.

关 键 词：同伴互评 互助学习 恶意评价 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]