Using Voting Approach for Event Extraction and Event-DCT, Event-Time Relation Identification

Temporal information extraction is a popular and interesting research field in the area of NaturalLanguage Processing (NLP) applications such as summarization, question answering (QA) andinformation extraction. In this paper, we have reported extraction of events and identification of differenttemporal relations between event-time and even-document creation time (DCT) within the TimeMLframework. Our long term plan is to make temporal structure that can be used in the applications likequestion answering, textual entailment, summarization etc. In our approach, we propose a voted approachfor (i) event extraction (ii) event – document creation time (DCT) relation identification (iii) event – timerelation identification from the text under the TempEval-2 framework. The contributions of this work aretwo-fold; initially features are extracted from the training corpus and used to train a CRF and SVMframework. Then, the proposal of a voted approach for event extraction, event-DCT and event-time relationidentification by combining the supervised classifiers such as Conditional Random Field (CRF) andSupport Vector Machine (SVM). In total we generate 20 models, 10 each with CRF and SVM, by varyingthe available features and/or feature templates. All these 20 models are then combined together into a finalsystem by defining appropriate voting scheme.