Spam Filter for Short Messages Using Winnow

Unsolicited bulk short message (namely spam message) is a major problem for mobile phone users. In this paper, an updatable spam filter embedded in mobile phone will be presented. Based on the traditional balanced Winnow algorithm, the system has been improved in many aspects which help to get better performance. Our spam filter is designed to process the raw text materials without time-consuming word segmentation for Chinese, and most importantly, to be easily extended to multi-class pattern by extending the original Winnow algorithm to multiple classifier. Detailed evaluations discussed in this paper will provide a strong support to the further research on spam filter for short messages.