期刊名称：International Journal of Engineering and Computer Science
印刷版ISSN：2319-7242
出版年度：2015
卷号：4
期号：7
页码：13512-13517
出版社：IJECS
摘要：Event extraction is a particularly challenging type of information extraction. Most current event extraction systems relyon local information at the phrase or sentence level. However, this local context may be insufficient to resolve ambiguity inidentifying particular types of events; information from a wider scope can serve to resolve some of this ambiguity. In this paper,we first investigate how to extract supervised and unsupervised features to improve a supervised baseline system. Then, wepresent two additional tasks to show the benefit of wider scope features in semi-supervised learning and active learning.Experiments show that using features from wider scope can not only aid a supervised local event extraction baseline system, butalso help the semi-supervised or active learning approach. The resulting efficient nugget pool is used to guide users’ exploration.Among the five stages of NMS framework, we pay our main attention on solving the technical challenges existed in nuggetcombination and refinement.A critical issue that makes nugget combination difficult is the distance metrics betweennugget (how can we know whether two nuggets are similar or not). For chunk refinement, trying to understand what a user islooking for when a nugget was generated is a difficult job which requires effective ”match” Cancer Attributes. In this thesis, wepresent KMSVM (K-Means Support Vector Machine) Classification solutions to both of these two challenges, and we haveconducted user study to carefully compare the performances of different distance metrics between nuggets.It is important to note that the training phase was done on 20% of the dataset, whereas the testing phase was done on theremaining 80% of the data set which are considered as unknown cases for the ALCs. The study proved that the best resultsobtained when the KMSVM select minimum reasonable number of features, while in the training phase the diagnostic accuracy is0.99 and the prognostic accuracy is 0.9, and the memories ALCs achieved in the testing phase a diagnostic accuracy 0.99 andanalytical accuracy 0.93.
