Deep learning-based information extraction

Abstract

Deep learning-based information extraction has shown great promise in automating the process of extracting structured information from unstructured data sources. This paper presents a literature review of deep learning principles and their application in information extraction tasks. Various approaches to information extraction have been proposed. A state-of-the-art method for universal information extraction and another subtask of information extraction, event argument extraction (EAE), is followed by this article. I understand and implement these models in-depth and employed exhaustive error analysis to find problems, in order to propose improvement methods. As a result, training with label noise is integrated and the performance of EAE has been improved on most evaluation metrics. Besides, this paper test and analyze how different prompts impact the outcomes. Last but not least, I verified few-shot ability for this task.