Towards Collaborative Spam Filtering Based on Collective Intelligence

It is important to make spam mail filters more intelligent.This paper proposes a collective intelligence-based approach to collaboratively build an unified knowledge by a cooperative multi-agent system (MAS), which is composed of a facilitating agent and a number of personal agents. The aim of this work is to support each personal agent to automatically filter spam mails from its mail box as referring to the centralized knowledge. The whole process is composed of two steps; i) personal agents can learn userpsilas actions by automatic feature extraction from the spams, and then, ii) they can communicate with the other agents to exchange the features. Due to domain-specific properties of the spam mail filtering, we have tried to formalize the features extracted from e-mails by an agent, so that it can be highly understandable and efficiently sharable with other agents. In particular, we have defined two types of features from the spam mails; i) field features, and ii) concept features based on keyphrases. Moreover, facilitator organizes hierarchical cluster structure to manage knowledge from these agents. Finally, we show the filtering performance of collaborative learning by comparing with personal agent.