Event Extraction from Unstructured Afan Oromo Text Using Deep Learning Approach

Abstract

Afan Oromo event extraction is to detect event instance(s) in texts, and if existing, identify the event as well as all of its participants and attributes. Event extraction is a challenging task in information extraction. Previous approaches highly depend on sophisticated feature engineering and complicated natural language processing (NLP) tools. In this study, we first come up with the language specific issue in Afan Oromo event extraction and then proposed a bidirectional LSTM model and LSTM language model to capture both sentence-level information without any hand craft features. We applied deep learning network for event extraction for Afan Oromo language model. We demonstrate that our approach is capable of detecting and extracting both events and arguments in Afan Oromo texts. Our proposed approach uses the rich neighboring context by taking both before and after word to be corrected through bidirectional input (i.e. left-to-right and right-to-left) deep LSTM network. Tensor flow deep learning python library was used to implement bidirectional LSTM algorithms. So to solve event extraction in Afan Oromo text, we employ deep neural networks with long-short term memory (LSTM) and bidirectional long-short term memory (BLSTM). For this study 3850, Afan Oromo sentences were used to train the system, which was semantically annotated with semantic roles using the BIO Tagging procedure and the PropBank annotation framework's principles. The experiment was conducted by using 90%, and 10% of the total dataset for training and testing respectively. Our techniques Bi-LSTM and LSTM achieved 84% and 81% accuracy of performance. So, the Bi-LSTM model outperforms the extraction of the system.