基于神经网络的卫生科技人员科研业绩评价

目的目前,关于科研评价的诸多方法中,多数从专家为评价指标赋权出发,没有完全摆脱人为主观因素的影响.本研究拟利用多层感知器(Conjugate Gradient Descent)算法对卫生科技人员的科研业绩进行综合评价.方法利用我校2002年拟晋升副高级职称专业技术职务(70人)和正高级职称专业技术职务的人员(48人)数据资料,结合科学计量学方法,拟定并筛选出能够反映科技人员既往和现今科研业绩的评价指标体系并科学地量化处理.然后利用Statistica Neural Networks软件,采用多层感知器(Conjugate Gradient Descent)算法进行数据训练.结果经过数百次数据训练,得出理想的网络.网络评价与实际同行评议结果的符合率分别达到90%和75%.结论只要确定了合理的评价指标体系并将之科学地转换为量化的变量值,同时具备足够的样本量,神经网络在科研业绩评价进而在其他科研评估研究与实践中是一种非常理想的方法.

Evaluation of Scientific Research Achievements of Health Care Professionals On the Basis of Artificial Neural Network

Objective Currently, many evaluation methods of scientific research achievements have been developed and applied.Majority of the methods start with setting the weights of evaluation indexes by authorities, thereby being fettered inevitably by a lot of subjective causatives. We planned to assess the scientific research accomplishments of health care professionals on the basis of neural networks.Methods We collected data of professionals from China Medical University who desired to promote associate professor (70 cases) and professor (48 cases) in 2002. Evaluation indexes were selected and quantified by scientometric methods and dealt with to adapting to neural network. The data were trained by Multilayer Perceptions ( Network Types) and Conjugate Gradient Descent (Training Algorithm) in Statistica Neural Networks software.Results Relative ideal network resulted through training hundreds of times. The consistency between actual peer review evaluation and neural network evaluation reached 90% (associate professor set) and 75% (professor set), respectively.Conclusion Artificial neural network was considered as an ideal approach to assess scientific research achievements and an available way in the other researches and practices of scientific research assessments, if only rational evaluation indexes were determined and quantified scientifically and the number of cases were adequate.