COMBINING TEXT EMBEDDING WITH ADDITIONAL KNOWLEDGE FOR INFORMATION EXTRACTION

Abstract

Information Extraction (IE) is an essential field of natural language processing (NLP). Over the years, researchers have studied numerous approaches and techniques to meet the challenges of different IE tasks. This dissertations explores various knowledge fusion techniques to combine diverse domain-independent and domain-specific knowledge with multiple text embedding techniques for effective IE. Specifically, this work presents a systematic investigation to combine different types of knowledge (e.g., lexical, syntactic, semantic, and domain knowledge) with different text embedding techniques (e.g., static and contextual embeddings) to achieve the state of the art performance in several IE tasks (e.g., Open IE, malware attribute identification and clinical relation extraction).