Information Technology Reference
In-Depth Information
of the data that arrives. As a consequence, the model built on old data, in
some cases inconsistent, becomes inappropriate to the new data. This problem
is known as conceptual drift and must be the main concern when synthesizing
a model from dynamic data [2]. Secker et al. [13] had already proposed the
use of an artificial immune system for this task, somehow comparable with the
perfomance produced by the na ıve Bayes approach.
7
Conclusions
In this paper we have proposed the application of a supervised antibody network
called SRABNET for spam filtering. Based on the use of a weighted index,
total cost ratio (TCR), to control the convergence window we obtained a better
performance with a robust network. Then, the use of SRABNET as a spam filter
instead of na ıve Bayes has shown to be an interesting choice for the user.
Even with the high accuracy of the antibody network, its performance can
be improved by adding some distinctive and domain specific features in the
representation as performed by Sahami et al. [5]. In [1] a comparison of meth-
ods for feature selection is presented, including information gain (IG), Mutual
Information (MI) or Chi squared (
χ
2
). We believe that the use of advanced fea-
ture selection techniques will accentuate the discriminant capability of filters for
spam.
Acknowledgment
The authors would like to thank CNPq and UOL (Universo On Line) for their
financial support.
References
1. O'Brien, C., Vogel, C.: Spam filters: Bayes vs. chi-squared; letters vs. words. In:
ISICT '03: Proceedings of the 1st international symposium on Information and
communication technologies, Trinity College Dublin (2003) 291-296
2. Tsymbal, A.: A case-based approach to spam filtering that can track concept drift.
Technical Report TCD-CS-2004-15, Trinity College Dublin (2004)
3. Cunningham, P., Nowlan, N., Delany, S.J., Haah, M.: A case-based approach to
spam filtering that can track concept drift. Technical Report TCD-CS-2003-16,
Trinity College Dublin (2003)
4. Androutsopoulos, I., Koutsias, J., Chandrinos, K.V., Spyropoulos, C.D.: An exper-
imental comparison of naıve Bayesian and keyword-based anti-spam filtering with
personal e-mail messages. Proceedings of the 23rd Annual International ACM SI-
GIR Conference on Research and Development in Information Retrieval (2000)
160-167
5. Sahami, M., Dumais, S., Heckerman, D., Horvitz, E.: A Bayesian approach to
filtering junk e-mail. In: AAAI-98 Workshop on Learning for Text Categorization.
(1998) 55-62
