垃圾邮件的概念漂移及过滤技术研究

Spam and spam filtering constitutes a game spam ecosystems. A long-term effective spam filtering technology should self-adaptive response to kinds of spam variations generated with time and user preferences, and which is known as concept drift in machine learning area. The dual-level concept drift detection algorithm was proposed to discern concept drift in which sustained misclassification was monitored when email classification. Regarding the spam concept scope changes caused by user preferences, the association and subjection strength algorithm was proposed between email fingerprints and concept subcategories based ontology. The proposed method was proved effective to handle the concept drift problem in spam filtering by comparing experiment.