Filtering Network Spam Message using Approximated Logistic Regression
Yiping Zheng, Fenglin Liu · Journal of Networks · 2014
The development of telecom network and Internet provides effective ways for communication. As an important way in communication, Short Messaging Service (SMS) via both telecom network and Internet has played an increasing important role in daily life. However, it usually suffers from spam SMS that causes misunderstanding and cheat. The highly varying content, network environment make the identification of spam message difficult. Although the previous methods to some extent can filter the spam messages, it usually fails to capture the semantic information because it simply relies on keywords. Thus, its accuracy is not satisfied enough. Also, their further applications to some difficult situations of spam SMS filtering are still limited by their shortcomings, i.e., their adaptation ability to network environment and their robustness to noise. Therefore, high efficiency spam SMS filtering method is of greatly important. In this paper, to overcome the shortcomings of previous methods for spam message filtering, we propose a new approach, linear discrimination based keyword selection with approximated logistic regression (KW-ALR). The proposed approach KW-ALR first extracts feature or keywords using linear discrimination analysis, and then trains spam recognition model based on approximated logistic regression over the extraced keywords. We evaluate the proposed approach KW-ALR over a standard data set SMS Spam Collection. The experimental result shows that our method KW-ALR for spam message filtering achieves higher accuracy over other methods.