A Study on Applying Support Vector Machines based Categorization Techniques to Identification of Chinese Spam Emails
| Autor: | Yuan-Ping Cheng, 鄭原平 |
|---|---|
| Rok vydání: | 2008 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 96 With the development and popularization of the internet in recent years, the use of the email tools has already become an inevitable trend. However, the usage also comes with a crucial problem due to the uncontrollable generation of spam emails. Such a side effect has been troubled the users of email systems. How to avoid receiving the unnecessary spam is becoming an important issue in current life. According to techniques related to text categorization using Support Vector Machines (SVMs) reporting in the literature, the SVMs based text classification approaches appear to have superior performance in general. Thus in this research we regard the emails as a kind of texts, and wish that we can utilize the SVMs techniques to effectively identify Chinese Spam emails. The research work is mainly using the SVMs techniques to compare the performance of identifying Chinese Spam emails, through considering the following aspects: the contents in different fields of the email, different feature selection methods (including choosing the noun only, choosing the verb only and choosing noun and verb), different feature selection strategies (including Binary, Term frequency (TF) and the product of the Term frequency multiply by Inverse Document Frequency(TF×IDF)), classifiers based on different kernel functions (including Linear SVMs classifiers, Gaussian RBF SVMs classifiers and Polynomial SVMs classifiers), different values of classifiers parameter C. Based on the factors mentioned above, we have been comparing experimental results from the experiments of automatic identification of Chinese Spam E-mails. Finally, we evaluated our experimental results by Recall, Precision, and F1 measures, in order to verify the performance difference of SVMs classifiers based on different selected features. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|