Combining Probability-Based Rankers for Action-Item Detection

摘要: This paper studies methods that automatically detect action-items in e-mail, an important category for assisting users identifying new tasks, tracking ongoing ones, and searching completed ones. Since consist of a short span text, classifiers can be built from document-level or sentence-level view. Rather than commit to either view, we adapt contextsensitive metaclassification framework this problem combine the rankings produced by different algorithms as well views. While is known work standard classification, its suitability fusing rankers has not been studied. In empirical evaluation, resulting approach yields improved are less sensitive training set variation, furthermore, theoretically-motivated reliability indicators introduce enable metaclassifier now applicable any where base used.