A Mail Filter with the Ability of Incremental-Learningbased on URL Analyzing

• Main Authors: Jing-Yao Huang, 黃敬堯
• Other Authors: Jyh-Jian Sheu
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/78580471950143963321

碩士 === 國立東華大學 === 數位知識管理碩士學位學程 === 98 === With the rapid development computer technology, Internet had become a part of human life in the past years. And e-mails were widely applied to advertisement. Because advertising cost was close to zero in the environment of Internet, “spam” was gathering serious. Spammer spread spam in order to increase visibility and search for business opportunities through advertising link. In this paper, we purpose a mail filter which was focused on analyzing of URL. Our mail filter was divided into three parts: (1) In Training phase, we constructed decision tree according to the features of training mails and gave each rule score. (2) In Execution phase, we extracted the features of unknown mail and scored this mail. We classified unknown mail into legitimate mail or spam according to its score. To prevent too many features would lead to confusion and reduce the false positive rate. We constructed the revising mechanism to re-compute mail’s final score according to the URL contained in the mail body. (3) In Incremental-learning phase, the filter would learn new URLs and adjust revising mechanism as unknown mails were inputted. According to the experiment, the accuracy of our mail filter was 99.64% and the recall reached 100%. The experiment result showed that our mail filter would successfully block spam.