基于改进平衡Winnow算法的短信过滤系统

Short message filtering system based on improved balanced Winnow

将黑白名单技术与Balanced Winnow算法相结合,实现对垃圾短信的过滤。采用CHI特征提取算法并对权重计算方法进行改进,同时提出了去除训练样本中野点的想法,通过判定去除野点,减缓在训练过程中出现的抖动现象。实验表明这种改进对于提高训练速度及提高短信过滤的性能均有很好的作用。

Spam short message（SMS） had been a big problem that disturbed cell-phone users and mobile operators recently.Combined black-white list technology and balanced Winnow algorithms to realize the aim of filtering spam short message,adopts CHI statistics feature selection algorithms in addition improve the calculating methods of feature＇s weight,meanwhile,put forward view of casting aside the outliers in training set,estimate and take out the outliers to reduce the jittery action at the training.Finally,experiments show that this amelioration improves the speed of training and profits performance of filtering.