基于框架方法的信用评估模型研究

发布时间：2018-12-24 10:06

【摘要】：人们的生活与信用的关联随着社会的进步及互联网的发展日益密切。由于数据的分散、评估方法标准的不一,目前信用体系包括影响指标体系、评估模型等还需要改善提高。论文针对信用数据指标落后、数据分散不利于融合以及评估方法欠缺特殊情形考虑的问题,对信用体系进行总结补充,将框架方法应用于信用影响指标体系的表示,并提出基于框架方法的信用评估方案。信用影响指标随着时代的进步潜在地发生着变化,会出现更多的因素会对信用造成影响。论文从政府信用、企业信用、个人信用以及针对农户群体的农户信用这几个方面分析补充了信用体系的可能性指标,并运用框架方法对其进行表示。由于框架方法结构清晰、便于扩展等优势,使得表示出的信用体系便于理解,且使用JSON格式直接存储这样的指标结构在内容补充变化时不会带来数据重构产生的重复性工作。结合Logistic回归模型对于各指标权重的训练结果,运用框架方法解决问题的一般思路,改进了匹配推理的原始方法。运用相似度计算进行实例匹配,且用概率差替换属性差进行加权平均。由于将数据转化为概率,因此,在用Logistic回归建模时,将特征项数据转化为对应的概率,以实现数据的一致性。通过将该方案与DT、SVM进行实验比较,表明了该方案在信用评估方面的可行性,且在数据样本较少时,提高了评估的准确度,另外,由于该方案是在粗糙搜索后进行实例匹配,极大地减少了时间的消耗。
[Abstract]:With the progress of society and the development of Internet, the relationship between people's life and credit becomes more and more close. Because of the dispersion of data and the different standards of evaluation methods, the current credit system, including the impact index system and the evaluation model, need to be improved and improved. Aiming at the problems that the credit data index is backward, the data dispersion is not conducive to the fusion and the evaluation method is not considered in the special situation, the paper summarizes and supplements the credit system, and applies the framework method to the expression of the credit impact index system. A scheme of credit evaluation based on frame method is proposed. With the progress of the times, more factors will influence the credit. In this paper, the possibility index of credit system is analyzed from the aspects of government credit, enterprise credit, personal credit and peasant household credit aiming at peasant household group, and the frame method is used to express it. Because the framework structure is clear and easy to extend, the expressed credit system is easy to understand, and the JSON format is used to store the index structure directly. Combined with the training result of Logistic regression model for each index weight, the original method of matching reasoning is improved by using the general idea of frame method to solve the problem. The similarity calculation is used for case matching, and the probability difference is used to replace the attribute difference for weighted average. Because the data is transformed into probability, when modeling with Logistic regression, the feature term data is transformed into the corresponding probability in order to realize the consistency of the data. By comparing the scheme with DT,SVM, it is shown that the scheme is feasible in credit evaluation, and the accuracy of evaluation is improved when the data sample is small. In addition, because the scheme is a case matching after rough search, The consumption of time has been greatly reduced.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O212.1

【参考文献】