文本分类支持向量机的i-ξα估计

i-ξα Estimator of SVM Text Classification

ξα估计是进行支持向量机模型选择的重要指标,它通过分析支持向量的特性,可以在训练一次的情况下估计出训练集发生"留一错误"的次数,进而判断当前模型参数选择的优劣。本文分析了文本向量及RBF核函数的特点,对用于文本分类领域的ξα估计进行了改进,提出了一种计算简便的"i-ξα估计"。实验表明,改进后"i-ξα估计"在保证准确性的前提下,明显提高了计算速度。

The ξα estimator is an important guideline for the SVM model selection . Using the characteristics of the support vectors, the ξα estimator can estimate the LOO errors with the SVM being trained for only one time, and then whether parameters are appropriate can be iudged. By analyzing the characteristics of text vectors and RBF kernel , the ξα estimator in text field is improved, which is called i-ξα estimator. Experimental results show the correctness of the i-ξα estimator is identical to that of the ξα estimator and its speed is much higher than the ξα estimator.