FASiL Adaptive Email Categorization System

摘要

This paper presents an adaptive email categorization method developed for the Active Information Management component of the EU FASiL project. The categorization strategy seeks to categorize new emails by learning user preferences, with a feature-balancing algorithm that improves the data training effectiveness and with a dynamic scheduling strategy that achieves the system adaptivity. The results of our evaluation with user-centric corpora constructed automatically from email servers are presented, with around 90% precision consistently being achieved after three months of use. Adaptivity of the system is also evaluated by studying system performance within the continuous three months.