이번에 리뷰할 논문은,

More Data, More Relations, More Context and More Openness: A Review and Outlook for Relation Extraction

이다.

#abstract

web text 및 인간의 지식이 확장됨에 따라, RE의 “more”이 필요해짐

1 Introduction

the real world is much more complicated than this simple setting:

(1) collecting high-quality human annotations is expensive and time-consuming,

• human annotation을 적게 또는 사용하지 않는 방향으로 -> Utilizing More Data

(2) many long-tail relations cannot provide large amounts of training examples,

• 길고 복잡한 관계 보완 → Handling More Complicated Context

(3) most facts are expressed by long context consisting of multiple sentences,

• 다수의 문장에 걸쳐 표현되는 관계들 → Performing More Efficient Learning

and moreover (4) using a pre-defined set to cover those relations with open-ended growth is difficult.

→Orienting More Open Domains

→ 이들을 보완하기 위한 further study 필요

2 Background and Existing Work

2.1 Pattern Extraction Models

2.2 Statistical Relation Extraction Models (SRE)

• feature-based methods
• kernel-based methods
• Graphical methods
• encoding text into low-dimensional semantic spaces and extracting relations from textual embeddings

2.3 Neural Relation Extraction Models (NRE)

NRE에 word embedding이 사용된다; position embeddings 등

• RNN
• CNN
• GNN
• attention-based neural networks
• Transformers, pre-trained language models