A new association rule-based text classifier algorithm
Supaporn Buddeewong, Worapoj Kreesuradej · 2005
This paper proposes a new association rule-based text classifier algorithm to improve the prediction accuracy of association rule-based classifier by categories (ARC-BC) algorithm. Unlike the previous algorithms, the proposed association rule generation algorithm constructs two types of frequent itemsets. The first frequent itemsets, i.e. L/sub k/ contain all term that have no an overlap with other categories. The second frequent itemsets, i.e. OL/sub k/ contain all features that have an overlap with other categories. In addition, this paper also proposes a new join operation for the second frequent itemsets. The experimental results are shown a good performance of the proposed classifier.