基于最大熵马尔科夫模型的绩效评价方法
Performance evaluation based on maximum entropy Markov model

本文提出了一种基于最大熵马尔科夫模型的绩效评价方法. 该方法采用马氏模型来定量化建模专家打分过程, 采用特征函数表征打分规则, 通过在训练集上最大化熵来获得符合专家经验的最优的打分模型. 与传统方法相比, 所提出的方法可以融合各种打分规则、专家经验和指标逻辑关系得到综合打分结果. 为了提高模型的训练和打分的效率, 本文提出了基于改进迭代算法的参数估计方法, 并利用Viterbi算法进行快速打分计算. 利用中国大洋协会绩效评价指标体系历史数据进行的仿真实验表明, 与BP神经网络方法和最大熵方法进行对比, 本文所提出的 方法具有更高的打分正确率.
This paper presents a new performance evaluation method based on the maximum entropy Markov model,which quantifies the process of scoring as a Markov process, represents the scoring rules by characteristic functions and obtains the optimal model parameters by maximizing the maximum entropy over a training sample set. Compared with other traditional methods, this method has the ability to combine complex scoring rules, expert experience with logical connection of the evaluated items to get comprehensive evaluation results. To improve the efficiency of training and scoring, this paper adopts the improved iterative scaling algorithm to obtain near-optimal model parameters and uses the Viterbi algorithm to quickly calculate the final evaluation results. The proposed method has been applied in the history data of China Ocean Mineral Resources R&D Association’s evaluation system for simulation. The experimental results show that this method has higher accuracy compared with BP networks and the classical maximum entropy model.