| In recent years, with the continued growth of the rapid development of mobile communications technology and the number of mobile phone users, send and receive SMS because it has at any time, inexpensive and convenient for people to convey information, etc., has become an important way of people’s daily life information exchange. Low criminals using SMS price, ease of mass and other characteristics of the user to send a large number of spam messages, seriously affecting people’s daily lives, including fraud class messaging also threaten the safety of people and property, so the spam SMS filtering technique needs improvement.Now spam SMS filtering technologies are: black and white list filtering, keyword filtering and content-based filtering. However, this single feature filtering capabilities filtration technology is limited, many spam messages are not filtered. Therefore, this paper constructs a list that contains black and white, keyword and content features such as intelligent spam filtering SMS filtering system. In text classification Naive Bayes classifier has a very critical condition, is to be classified text features are independent of each attribute, but this condition can not be satisfied in some occasions. Naive Bayes check rates lower full text classification, easy to normal SMS messages classified as spam, inconsistent with the actual expectations of the people. For the above two issues, this paper uses an improved Bayesian classification algorithm, it uses an improved method of class conditional probability estimates and improved discriminant function independently of each other to solve the problem of low and recall. The two classification algorithms experimental comparison, the results show that the improved Bayesian recall and overall performance is better than Naive Bayes.With the research and development of SMS spam filtering system, criminals also came up with all sorts of means, by changing the content of messages in an attempt to evade interception filtration system. For message content transformation include: Adding interference character, complex characters replace, split, replace the word, homophonic word replacement. These means endless spam SMS filtering system to a huge challenge. This paper conducted a study of these transformations, and gives a solution. The main work of this paper include:1. Text analysis and comparison of the advantages and disadvantages of different classification algorithms, according SMS feature selection classification Bayesian classification algorithm as research methods.2. Solve the conversion problem of spam messages, such as: Traditional replace, homophonic substitutions, splits the word replace, inter-symbol interference, etc.3. For five feature extraction methods were compared using an integrated word frequency and the advantages of mutual information feature extraction method and mutual information feature extraction methods and conducted experiments comparing.4. Reference existing spam SMS filtering technology, the integration of black and white list filtering, keyword filtering and content-based filtering, constructs a Bayesian classification algorithm based on improved junk SMS filtering system, and experimental analysis of its performance. |
