摘要
KⅢ模型在模式识别中表现出良好的性能,但它只能对样本给出与否的预测。TCM是基于算法随机性理论提出的一种分类算法,不仅可以判断出分类类别,同时还可以对每个预测结果给出可靠性的度量。本文把直推式信任机器(TransductiveConfidence Machine,TCM)跟KⅢ模型结合,TCM-KⅢ可以有KⅢ模型的预测能力,还可以给出可靠性度量,丰富了输出信息。文本分类一直是一个活跃的课题,所以本文把TCM-KⅢ应用在分本分类中,实验表明TCM-KⅢ和单一的KⅢ模型预测的准确率相近,但给出了可靠性度量,可以进行有效的风险控制。
The KⅢmodel in pattern recognition show good performance,but it can not give an indication of how "good" the predictions are.TCM is based on algorithmic theory of randomness,not only determine the classification categories,but also can predict the results for each measure of reliability is given.This thesis combined transductive confidence machine with the KⅢmodel,TCM-KⅢnot only has KⅢability to forecast,but also gives the reliability of measurement.TCM-KⅢcan be effective risk control.Text classification has been an active topic,this article applied to the TCM-KIII points of the classification,the results show that TCM-KⅢand single KⅢmodel prediction accuracy is similar,but given the reliability of the measure.
出处
《微计算机信息》
2012年第2期151-152,55,共3页
Control & Automation