北京市企业标准化评审结果分析方法
Analysis method for standardization reviews on Beijing enterprises

北京市企业标准化评审后留下了一万多家企业百万余条评分结果，在海量数据的支撑下，可以采用大数据分析手段来探索各扣分项扣分频次之间的相关关系。目前最常用的方法是关联规则挖掘，然而关联规则挖掘只能给出定性的相关关系，无法在定量方面给出结论，使得对数据的信息挖掘停留在定性的水平。神经网络作为另外一种广泛应用的数据挖掘方法能够有效拟合复杂的非线性关系，但是在数据挖掘中，神经网络在输入输出选择上存在很高的试错成本。该文将关联规则挖掘与神经网络方法结合使用，首先用关联规则挖掘筛选出扣分项之间的关联规则逻辑，然后将选出的关联规则作为神经网络的输入与输出进行训练，找到了18项扣分项之中的3项与其他8项之间的强相关关系，神经网络预测值与实际值之间拟合直线的相关系数达到了0.84以上。实验结果表明：该方法可以实现对企业扣分项的相关关系挖掘，并可以将结果用于扣分频次预测。
Abstract：There are over one million items from ten thousand enterprises in the Beijing Enterprise Standardization Review. Big data analytical methods can be used to analyze the relationships between the deduction counts of the review items because of the large data volume. The most popular method is association rules, but these are qualitative, not quantitative. Neural network, another widely used data mining method, are able to solve complex non-linear problems but requires much effort to choose the suitable inputs and targets. This article combines these two methods with the association rules used to select the inputs and targets from the review items and the neural network used to relate the inputs and the targets. A test gave a strong correlation between 3 selected review items and 8 other review items with a correlation coefficient of the fitting curve of over 0.84 between the predicted targets and the real value. Thus, this combined method can improve data mining of the enterprises review items with the result used to predict the deduction counts of the selected items.