Similarity-based agents for e-mail mining

With Internet use continuing to explode, and due to the simplicity of sending e-mails to many people, recent years have seen the time spent in dealing with unnecessary and irrelevant e-mails increasing. In general, we note that the efforts of the scientific and industrial communities have been focused on the idea of smart filtering services. Our approach is different: the user wishes to send an e-mail to an appropriate reader, i.e. a user whose "profile" is compatible with the content of the e-mail itself. The profile is described in terms of topics that are related to the e-mail argument through a similarity-based network. The e-mail writer establishes this cognitive frame on the client-side, exploiting similarity-based reasoning. Then a search engine, based on mobile computation, is triggered: a number of autonomous agents are created and sent on to the network. The agents work as a Web-crawling spider, not exploring the net indiscriminately but searching domain-relevant documents directly on potential reader hosts. From this kind of document, the agent extracts logic-based knowledge that is processed by the similarity deduction engine. As a result, the agent returns an evaluation of the users' degree of interest in receiving the potential e-mail. On the client-side, a collector receives the different evaluations in order to define the final user mailing list by means of a flexible mechanism.