基于多任务学习的邮件过滤系统的研究  

Research of Spam Filter System Based on Multitask Learning

作　　者：许棣华[1,2] 王志坚[1] 

机构地区：[1]河海大学计算机信息与工程学院,江苏南京210098 [2]南京邮电大学计算机学院,江苏南京210003

出　　处：《计算机技术与发展》2010年第10期137-140,共4页Computer Technology and Development

基　　金：国家自然科学基金(60805022);国家高技术研究发展计划(863)(2007AA01Z178);南京邮电大学青兰计划(NY206034)

摘　　要：随着电子邮件的广泛使用,如何有效地避免和防范垃圾邮件的侵扰已成为一个亟待解决的问题。受机器学习在邮件过滤中研究和应用的启发,利用多任务学习(multitask learning)的特性,将判断一个用户的邮件是否为垃圾邮件看作一个任务(task),利用多任务学习中任务相关性假设,提出一种基于多任务学习的邮件过滤系统。实验表明,该系统对中英文邮件语料都是可靠和有效的,尤其对于同一邮件列表(mail list)中的用户的邮件。With the widespread use of e-mail,how to effectively avoid and prevent junk e-mail has become very urgent.Inspired by the research and application of machine learning in spam filter,a spam filter based on multitask learning is proposed,considering whether a user＇s e-mail is spam or legitimate as a task.Using tasks relevance coefficient,the system classifies emails,with the assumption of task relevance in multitask learning.Experiments show that the system is reliable and effective for both English and Chinese corpus,especially for the mails in a mail list.

关 键 词：多任务学习 任务相关性 邮件过滤 分类 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]