A Categorization Algorithm for Harmful Text Information Filtering

摘要

Harmful text information filtering is a typical pattern recognition problem of small sample, the prediction result of classifier was biased towards the class with more samples, because of the samples that including the harmful information were difficult to gain. Construct virtual samples is an effective means to solve the problem of pattern recognition in the small sample, using the up-sampling method to construct virtual samples in the data layer, the traditional KNN algorithm has been improved: a small sample set is divided into clusters by using the K-means clustering, the virtual samples are generated and verified the validity in the cluster. The experimental results show that this method can construct the virtual samples which are similar to the real sample characteristics, and expand the small sample collection in order to effectively identify the harmful text information.