Research On Key Techniques Of Spam Short Message Filtering

Recently, the inundant spam short messages have become a problem which suffers the mobile service provides and normal uses. This article addresses on the problem of recognizing spam short messages in spam filtering, and implements the Bayes methods, SVM methods and social network methods with the constraint of the short messages sending system to research the two key problems in spam short messages filtering, which are the individual spam short message contents recognition and individual spam short message originator recognition. The main work includes the following:1. Unified Spam short message filer modelA unified spam short messages filter model is proposed which can deal with varied SMS sending methods. This model considers the detailed characteristics of different SMS sending methods and makes full use of the mature short messages blocking techniques, which can block the spam short messages by contents and block the spam short message by originators. It could reduce the cost when the new spam short message recognition methods are applied .2. Multi-Characteristic based spam short messages recognition algorithmA multi-Characteristic based spam short messages recognition algorithm is presented, which combines the behavior characteristics of short message originators and the text characteristics of short message, and utilizes Bayes classifier and SVM classifier to learn and classify the spam short messages. To provide the incremental learning, a series of polices of self-feedback has been proposed in this algorithm. The experiments show the algorithm could promote both in the efficiency and accuracy comparing with the traditional methods.3. Social network based spam short message originators recognition algorithmA formal model for the short message sending network in social network has been built. And several experiments are carried out to prove the properties of the short message sending network, such as small world, and scale-free etc. Further work has been done to analyze the abnormal pattern and behavior of spam short messages. Then we propose a voice connection based associating filter algorithm NASFA, the experiments show our algorithm can recognize the spam short message originators accurately and decrease the wrong recognition ratio.4. Location based spam short message originator recognition algorithmAs the spam short messages have notable spatial features, we present a spam short message originator recognition algorithm based on its base station transceiver location code, cell id and content features.5. Quick matching filter algorithm for short message gateway originated short messagesAimed to the characteristics of short message gateway originated short messages, i.e. fast batch sending speed, wide range of the sending frequency, and without social characters and few classifying characters, we propose a quick matching based filter algorithm which can deal with the spam short messages originated from the INTERNET via the short message gateway successfully. It utilizes the preprocessing mechanism in message blocks to avoid the anti-filtering tricks and filter the messages by the dynamic total messages sent and message contents. And a keywords weighted method has been introduced to adjust the maximum number allowed to send based on the weights.Finally, a recapitulative conclusion is given, and the future research directions are proposed.