[Savannah-hackers] submission of Distributed Message Classifier - savannah.nongnu.org

[devel]

A package was submitted to savannah.nongnu.org
This mail was sent to address@hidden, address@hidden


Santosh Dawara <address@hidden> described the package as follows:
License: gpl
Other License: 
Package: Distributed Message Classifier
System name: dmsgclass
Type: non-GNU

Description:
Distributed Bayes is an attempt to develop a Distributed Bayesian email 
classification system. Potential applications include the filtering of Spam. 
The advantage of Distributing the problem of classification allows the 
possibility of sharing conditional probabilities derived by individual email 
agents. This is extremely critical if we want to be able to quickly nail and 
filter adapting commerical junk mail. 

Agents could conceptually share email Templates or just particular features and 
their conditional probabilities. For example, after training on an email box, 
an Agent comes up with the conclusion that \"Save More\" indicates that the 
message is Spam with a 90% probability. This rule could then be shared with 
other Agents who won\'t have to be trained to discover the same rule. 

There is a flip side of course. There are exceptional conditions. The 
usefulness of the agent depends on how well we can handle these exceptional 
conditions. With concepts in Distributed Artificial Intelligence, we could 
create a hierarchy of rules, classifying them as local and global. Local rules 
could inhibit global rules (or beliefs). The point is, there are several 
interesting problems that require investigation.