LSTM-based Model for Extracting Temporal Relations from Korean Text

Abstract

Temporal information extraction plays an important role in providing a Q&A service or an interactive system that can grasp the user's intention and context of a conversation. It is particularly difficult to correctly recognize the temporal relations from Korean text owing to the inherent linguistic characteristics of the Korean language. In this paper, we propose a deep neural network designed to capture the temporal context from Korean natural language sentences based on long short-term memory (LSTM) for extracting the relationships among the time expressions and events. There are three types of temporal information extraction: TIMEX3, EVENT, and TLINK extraction; however, we only aim to extract TLINKs (i.e., temporal relations) between TIMEX3 and EVENT entities that have already been extracted from the given sentences. We also demonstrate the performance of our LSTM-based model when extracting temporal relationships from human-annotated datasets.