Polibits, Vol. 46, pp. 55-59, 2012.
Abstract: Event extraction is a popular and interesting research field in the area of Natural Language Processing (NLP). In this paper, we propose a hybrid approach for event extraction within the TimeML framework. Initially, we develop a machine learning based system based on Conditional Random Field (CRF). But most of the deverbal event nouns are not correctly identified by this machine learning approach. From this observation, we came up with a hybrid approach where we introduce several strategies in conjunction with machine learning. These strategies are based on semantic role-labeling, WordNet and handcrafted rules. Evaluation results on the TempEval-2010 datasets yield the precision, recall and F-measure values of approximately 93.00%, 96.00% and 94.47%, respectively. This is approximately 12% higher F-measure in comparison to the best performing system of SemEval-2010.
Keywords: About event, TimeML, Conditional Random Field, TempEval-2010, WordNet
PDF: A Hybrid Approach for Event Extraction
PDF: A Hybrid Approach for Event Extraction