Detection Method of Blog Spam Based on Categorization and Time Series Information
Toshio TERAGUCHI, Kenji D. Nakamura, Shigenori Tanaka, Koichi Kitano · 2012
Recently, a blog is well known as a tool for transmitting information easily. However, as the blog spam increases, the method of filtering spam is required to be efficient. In early researches, the method for detecting spam mails with Bayesian filter detects the spam from the characteristics of spam words appearing in spam mails with a high degree of accuracy. How-ever, there are some problems in applying the Bayesian filter to the blog spam detection. First, it takes a lot of man-hours to keep high accuracy continuously. Second, the accuracy of spam detection decreases because there are too many various words in a blog and the relative number of word occurrences de-creases. Furthermore, we have to consider the time when each word occurred. Therefore, in this paper, we acquire information to update the judgment information automatically and calculate spam probability of words with every category to cope with these problems. In addition, we use the time-series information to revise spam probability of words to cope with the problem that the words that occur change over time. With these countermeasures, we propose a method for detecting a new blog spam. With comparative experiments, the present method is better adapted to any existing method.