Breaking the Closed World Assumption in Text Classification

Abstract

Existing research on multiclass text classific a-tion mostly makes the closed world assum p-tion, which focus es on designing accurate classifiers under the assumption that all test classes are known at training time. A more r e-alistic scenario is to expect unseen classes dur ing testing ( open world ) . In this case, the goal is to design a learning system that class i-fies documents of the known classes into their respective cl asses and also to reject doc u-ments from unknown classes . This problem is called open ( world ) classification . This paper approach es the problem by reducing the open space risk while balancing the empirical risk . It proposes to use a new learn ing strategy, called c enter - based similarity (CBS) space learning (or CBS learning ) , to provide a novel solution to the problem . Extensive e xper i-ments across two datasets show that CBS learning give s promising results on multiclass open text classification compared to state - of - the - art baselines.