The Solution of Xiaomi AI Lab to the 2021 Language and Intelligence Challenge: Multi-format Information Extraction Task

Abstract

Information Extraction is a challenging task in Natural Language Processing. There are several formats of information extraction that researchers mainly focused on. Typically, relation extraction and event extraction are two type of information extraction tasks that could facilitate many related business. In this work, we will introduce our solutions for the 2021 Language and Intelligence Challenge. We propose our methods for the relation extraction and event extraction tasks, respectively. The event extraction task is further divided into sentence level and document level. In relation extraction task, we propose a knowledge-based relation extraction method; in event extraction task, we propose a hybrid method which utilize event types and event triggers separately to extract event roles and event arguments. Finally, our solution ranked the second in the private leaderboard.