LAI: Label Annotation Interaction based Representation Ehancement for End to End Relation Extraction

Anonymous

Abstract: While numerous studies on end-to-end relation extraction (E2ERE) have centered on enhancing span representations to improve model performance, challenges remain due to the gaps between subtasks (named entity recognition and relation extraction) and the modeling discrepancies between entities and relations. In this paper, we propose a novel Label Annotation Interaction based representation enhancement method for E2ERE, which institutes a two-phase semantic interaction to augment representations. Specifically, we firstly feed label annotations that are easy to manually annotate into a language model, and conduct the first round interaction between three types of tokens with a partial attention mechanism; Then construct a latent multi-view graph to capture various possible links between label and entity (pair) nodes, facilitating the second round interaction between entities and labels. A regimen of rigorous experimentation demonstrates that LAI-Net achieving performance parity with the current SOTA models on ADE/SciERC dataset in terms of NER task (a SOTA performance has been achieved on the ACE05 dataset pecifically), and establishing a new SOTA result (with nearly a 10% advance on the SciERC dataset for RE specifically) in terms of RE task.

Paper Type: long

Research Area: Information Extraction

Contribution Types: NLP engineering experiment

Languages Studied: English