本文選題：垃圾郵件　切入點：樸素貝葉斯　出處：《西安電子科技大學(xué)》2012年碩士論文

【摘要】：垃圾郵件過濾是當(dāng)前互聯(lián)網(wǎng)應(yīng)用中急需解決的一個重要課題，日益受到人們的關(guān)注。一般而言，垃圾郵件是指同一個發(fā)件人在同一時間將同一電子郵件寄往許許多多不同的用戶，它的內(nèi)容主要包括廣告和一些政治宣傳信件。如果經(jīng)常收到這樣的電子郵件，就會令人感到十分厭煩，而大量的信件更會擾亂電子郵件的正常使用。垃圾郵件過濾實際上是一類文本分類問題，樸素貝葉斯分類器是其中一種簡單而有效的分類方法。該方法的不足在于它假定所有的屬性是互相獨立的，往往無法滿足實際應(yīng)用。但是，如果不作條件獨立性假設(shè)，必然會導(dǎo)致組合爆炸。為此基于改進貝葉斯的垃圾郵件過濾算法已受到了越來越多研究人員的廣泛關(guān)注。 本文首先研究了垃圾郵件過濾方法和和相應(yīng)過濾算法，比較了一些典型算法的優(yōu)缺點，研究了電子郵件的發(fā)送和接收協(xié)議，分析了當(dāng)前垃圾郵件過濾技術(shù)研究現(xiàn)狀；根據(jù)電子郵件系統(tǒng)的工作原理，著重分析了基于貝葉斯網(wǎng)絡(luò)的垃圾郵件過濾技術(shù)；結(jié)合相應(yīng)實例，分析了樸素貝葉斯郵件分類的分類特點及精度。提出了客戶端郵件過濾軟件的缺失問題，基于此設(shè)計了一個客戶端郵件過濾系統(tǒng)。最后基于VSTO結(jié)合Outlook實現(xiàn)了一個郵件自動過濾系統(tǒng)。該系統(tǒng)集成了手動規(guī)則、黑名單、白名單、自動規(guī)則、單機器學(xué)習(xí)過濾器、集成學(xué)習(xí)過濾器等多種過濾手段，主要在計算機客戶端對收到的新郵件進行處理，使用已經(jīng)被分類為垃圾郵件和合法郵件的郵件作為實驗來源，并獲得了相應(yīng)的特征模式。再對該特征進行學(xué)習(xí)，從而實現(xiàn)過濾的目的。 經(jīng)測試，表明該系統(tǒng)功能齊全，過濾效果十分良好，其查準(zhǔn)率≥95%，誤拒率≤2%，誤收率≤10%，，具有很高的推廣價值。其次，該系統(tǒng)還可以作為Outlook的過濾插件使用，能對Outlook收件箱中的郵件進行自動郵件過濾。 反垃圾郵件的精確度和查全率一直是垃圾郵件過濾系統(tǒng)所要研究的重要方向，在以后的工作中還要繼續(xù)在這方面加大研究力度，不斷提高反垃圾水平。
[Abstract]:Spam filtering is an important problem that needs to be solved in the current Internet application, and has been paid more and more attention.Generally speaking, spam means that the same sender sends the same email to many different users at the same time. Its contents mainly include advertisements and some political propaganda letters.It can be tiresome to receive such emails on a regular basis, and a large number of them can disrupt their normal use.Spam filtering is actually a kind of text classification problem, and naive Bayes classifier is one of the simple and effective classification methods.The disadvantage of this method is that it assumes that all attributes are independent of each other and can not satisfy the practical application.However, if the hypothesis of conditional independence is not made, it will inevitably lead to a combination explosion.Therefore, the improved Bayesian spam filtering algorithm has attracted more and more researchers' attention.Firstly, this paper studies the spam filtering methods and corresponding filtering algorithms, compares the advantages and disadvantages of some typical algorithms, studies the sending and receiving protocols of email, and analyzes the current research status of spam filtering technology.According to the working principle of email system, the spam filtering technology based on Bayesian network is analyzed, and the classification characteristics and accuracy of naive Bayesian mail classification are analyzed.This paper puts forward the lack of client mail filtering software, and designs a client mail filtering system based on this.Finally, an automatic mail filtering system based on VSTO and Outlook is implemented.The system integrates manual rules, blacklists, white lists, automatic rules, single-machine learning filters, integrated learning filters, and so on.The spam and legitimate mail are used as the experimental sources and the corresponding characteristic patterns are obtained.Then the feature is studied to achieve the purpose of filtering.The test results show that the system has complete function and very good filtration effect. Its precision ratio 鈮

本文編號：1709776