Zhemin Zhu (2015-03-24 13:45 - 14:30 in ZI-2042)
We focus on two sub-tasks of open relation extraction. (i) Extracting relation expressions from dependency paths. We propose a novel probabilistic simplification model SimpleTRE, which learns and applies typical simplification operations, including insertion, deletion, reordering, splitting, and substitution, to dependency paths to extract simple, informative and coherent relation expressions. (ii) Identifying relevant temporal expressions. This task is to associate temporal expressions in a sentence to relations extracted from this sentence. We develop a SVM classifier using dependency features for this task. A Wikipedia dataset is manually annotated for training and testing. Experiments show on both tasks our model obtains better F1 than the state-of-the-art relation extraction system on our Wikipedia dataset.