Abstract: We design a novel interactive learning algorithm that directly incorporates expert knowledge into WSD model training process. In the new learning algorithm, experts can express knowledge as the association between senses and textual patterns, and the machine learning algorithm can directly learn from such association. Experiments on one biomedical literature data set and two clinical notes data sets show that the proposed algorithm makes better use of human experts in training WSD models than all previous approaches, achieving the state-of-the-art performance with least effort from domain experts.
Learning Objective 1: Learning the state-of-the-art interactive learning algorithm for medical word sense disambiguation that requires minimal human efforts.
Learning Objective 2: Getting familiar with the cold-start problem data-driven methods and ideas of leveraging prior knowledge to alleviate this problem.
Yue Wang (Presenter)
University of Michigan
Kai Zheng, University of California, Irvine
Hua Xu, University of Texas Health Science Center
Qiaozhu Mei, University of Michigan