UniRE: A Unified Label Space for Entity Relation Extraction

Abstract

Many joint entity relation extraction models setup two separated label spaces for the two sub-tasks (i.e., entity detection and relation classification). We argue that this setting may hinder the information interaction between entities and relations. In this work, we propose to eliminate the different treatment on the two sub-tasks' label spaces. The input of our model is a table containing all word pairs from a sentence. Entities and relations are represented by squares and rectangles in the table. We apply a unified classifier to predict each cell's label, which unifies the learning of two sub-tasks. For testing, an effective (yet fast) approximate decoder is proposed for finding squares and rectangles from tables. Experiments on three benchmarks (ACE04, ACE05, SciERC) show that, using only half the number of parameters, our model achieves competitive accuracy with the best extractor, and is faster.

OneRel: Joint Entity and Relation Extraction with One Module in One Step

“…The second class is multi-module one-step, which extracts entities and relations in parallel, and then combines them into triples. For example, Miwa and Bansal (2016); Zhang, Zhang, and Fu (2017); Wang et al (2020Wang et al ( , 2021 treat entity recognition and relation classification as a table-filling problem, where each entry represents the interaction between two individual words. Sui et al (2020) formulate the joint extraction task as a set prediction problem, avoiding considering the prediction order of multiple triples.…”

“…Although promising, this kind of models suffers from the problem of cascading errors since mistakes in early steps may affect the prediction results of later steps. The second category attempts to identify entities and relations separately, and then combine them into triples based on their latent correlations (Wang et al 2020;Sui et al 2020;Wang et al 2021). However, due to insufficient mutual constraints between entities and relations in the separate recognition process, such methods tend to produce redundant information, leading to errors when assembling triples (Zheng et al 2017).…”

Towards Effective Multi-Task Interaction for Entity-Relation Extraction: A Unified Framework with Selection Recurrent Network

“…To address this, recent works seek to jointly solve the extraction of entities and relations. Some of them (Miwa and Sasaki, 2014;Gupta et al, 2016;Zhang et al, 2017;Tran and Kavuluru, 2019;Wang and Lu, 2020;Wang et al, 2020bWang et al, , 2021 cast NER and RE as a table filling problem. They attempt to capture the implicit interaction between NER and RE by sharing the same encoder and weights for simultaneous prediction of entities and relations.…”

“…Current approaches aim to jointly solve the two sub-tasks to take advantage of the benefits of their inter-task correlation. Specifically, some of them cast NER and RE as a joint table filling problem (Miwa and Sasaki, 2014;Gupta et al, 2016;Zhang et al, 2017;Tran and Kavuluru, 2019;Wang and Lu, 2020;Wang et al, 2020bWang et al, , 2021. This allows NER and RE to be performed in one stage through implicit multi-task interaction, realized by shared feature space.…”

CompactIE: Compact Facts in Open Information Extraction

“…Existing neural systems adopt a sequence labeling (Kolluru et al, 2020a;Wang et al, 2021;Ro et al, 2020) or a sequence generation (Kolluru et al, 2020b) approach to identify triples and their constituents, typically all at once, or through a pipeline that first identifies the relations and then their corresponding arguments. None of these methods guarantee that the extracted triples will be compact and share constituents.…”