Project by Fidelis Assis won the first three awards among 32 spam filters
| The anti-spam filter program developed by telecommunications engineer Fidelis Assis, technical manager of Embratel?s Internet Support Service, won the 15th TREC Spam Track 2006, the most important international conference focused on evaluation of current and proposed spam filtering approaches. Promoted by the National Institute of Standards and Technology (NIST), a US Commerce Department agency, and by the US Department of Defense, the competition gathered 9 teams from 7 countries participating with 32 filters. |
Fidelis Assis
|
The anti-spam filter program of the Brazilian candidate competed with projects presented by universities, technology institutes and research laboratories from the US, Germany, Canada, China and other countries. All the teams could submit up to four anti-spam filters. The four variants of OSBF-Lua were rated first, second, third and fifth. The US Tufts University was rated fourth.
Intelligent System
The OSBF-Lua filter has a feedback mechanism which allows the user to train it on what is and what is not a spam. Therefore, it learns the user profile from the user himself, and in a short time it gets accurate enough to automatically select the ?good? and "bad" messages. The combination of theses techniques produces a quick, light, highly accurate and intelligent filter, making unnecessary the frequent update of anti-spam version or rules.
The OSBF-Lua filter is currently used to protect the emailing services of Embratel Group companies, including Click 21 Internet support service, Embratel?s free Internet portal, Star One, Embratel Group?s satellite arm, etc. "The program, which is available on a free basis, can be downloaded from http://osbf-lua.luaforge.net/."
Fidelis Assis has been working for Embratel for 32 years now. He has a telecommunications engineering degree and a masters degree in computing science with IME (Instituto Militar de Engenharia), and started to develop the project in 2004 as an alternative to heavy, complex filters. At that time, he developed a text characteristics extraction technique, consisting of assessing key words and/or terms characteristic of a spam message called OSB. Next, he developed a method to select the most significant characteristics among the extracted ones (OSBF - Orthogonal Sparse Bigrams with confidence Factor), which is used with the OSBF-Lua filter.
The official result of the 15th TREC Spam Track 2006 will be published in March.
|
