現今e-mail為人們傳遞訊息的主要媒介，但e-mail帶來便利的同時，也衍生了許多安全性的問題，如盜用帳號、竊取資料等網路犯罪事件層出不窮。因此迫切需要能有效鑑定e-mail來源是否可靠的方法。 作者鑑定(Authorship Identification)為根據文章的寫作風格特徵(style features)，而提供最有可能的作者之方法。應用於e-mail上，則可藉由判別可疑e-mail之寫作風格特徵，得知是否出自於原作者。但目前針對e-mail作者鑑定的研究並不多，所提出的方法皆有效率低落的缺點，甚至有些方法只能適用於特定情境之下，因而有適用性差的缺點。 本研究所提出的UserProtector 演算法，為同時兼顧高效率和高適用性的方法。由於其他研究所的方法皆須對整封e-mail進行掃描，而UserProtector只需對信件標題掃描即可，因此具有效率高的優點。另外，因採用Character n-grams作衍伸的方法萃取風格特徵，因此各種情境下皆能有效萃取風格特徵，因此具有適用性高的優點。 Nowadays, people use E-mail as the main media to transfer messages. However, while e-mail is convenience for people, it also brings out many problems of security. Internet crimes like account usurping, data stealing are getting worse. Therefore, a method to identify e-mail sources efficiently is urgently necessary. Authorship Identification can base on the style features of articles to provide the most possible writers. It can be used to identify the original writer by judging dubious style features of an e-mail. But, there aren’t many researches that focus on identifying e-mail writers right now. They all have a chief defect of low efficiency. Moreover, some of them can be only used in specific circumstances. Hence, a defect of low suitability appears as well. To take both high efficiency and suitability into consideration, this research provides an algorithm: UserProtector. Duo to other methods need to scan all content of one e-mail, UserProtector only scan the e-mail subject. Consequently, it has an advantage of high efficiency. Further, by evolving Character n-grams to extract style features, every kind of circumstances can be extracted style features efficiently. For this reason, it has an advantage of high suitability.
