Automated generation of ham rules for Vietnamese spam filtering

The topic of spam filtering has been thoroughly studied by researchers in the past few decades. There has been successful works with high spam detection rates, yet no paper has described a method which can effectively detect spam and, at the same time, measure the importance of ham emails. In this paper, the authors propose a method of generating SpamAssassin rules which can indicate the degree of importance of an email message. Specifically we added a proportion of negatively weighted ham rules and adapted HPSOWM, an efficient evolutionary algorithm, to optimize SpamAssassin rule scores. As a result, using our new rule set, SpamAssassin is able to give indicative scores for both spam and ham. These scores can be utilized by email clients to categorize incoming messages based on their importance to user. Various experiments were conducted to evaluate our method. In addition, a conclusion was drawn about the best ratio of spam rules and ham rules.