Abstract

Distance supervision is widely used in relation extraction tasks, particularly when large-scale manual annotations are virtually impossible to conduct. Although Distantly Supervised Relation Extraction (DSRE) benefits from automatic labelling, it suffers from serious mislabelling issues, i.e. some or all of the instances for an entity pair (head and tail entities) do not express the labelled relation. In this paper, we propose a novel model that employs a collaborative curriculum learning framework to reduce the effects of mislabelled data. Specifically, we firstly propose an internal self-attention mechanism between the convolution operations in convolutional neural networks (CNNs) to learn a better sentence representation from the noisy inputs. Then we define two sentence selection models as two relation extractors in order to collaboratively learn and regularise each other under a curriculum scheme to alleviate noisy effects, where the curriculum could be constructed by conflicts or small loss. Finally, experiments are conducted on a widely-used public dataset and the results indicate that the proposed model significantly outperforms baselines including the state-of-the-art in terms of P@N and PR curve metrics, thus evidencing its capability of reducing noisy effects for DSRE.

Highlights

  • Relation Extraction (RE) is vital for NLP tasks such as information extraction, question answering and knowledge base completion

  • The main contributions are summarised as: (1) We make the first attempt to use the concept of curriculum learning for denoising Distantly Supervised Relation Extraction (DSRE) and present a novel collaborative curriculum learning model to alleviate the effects of noisy sentences in an entity pair bag

  • As with the evaluation metrics used in the literature, we report our results using Precision-Recall curve (PR-curve) and Precision at N (P@N) metrics

Read more

Summary

Introduction

Relation Extraction (RE) is vital for NLP tasks such as information extraction, question answering and knowledge base completion. The curriculum learning training method for DSRE is used with the assumption that entity pair bags contain corrupted labelled sentences, which are difficult components to learn in the curriculum. The main contributions are summarised as: (1) We make the first attempt to use the concept of curriculum learning for denoising DSRE and present a novel collaborative curriculum learning model to alleviate the effects of noisy sentences in an entity pair bag. In this model, we define two collaborative relation extractors to regularize each other and boost the model’s learning capability. Instead of using a separated complex noisy sentence filter and two-step training in baseline models, our model can alleviate noise effects during a single training and is easy to implement. (3) We are the first to apply an internal CNNs self-attention mechanism to enhance a multilayer CNNs model for DSRE. (4) We conduct thorough experiments on the widely-used NYT dataset, and achieve significant improvements over state-of-the-art models

Related Work
Methodology
Inputs
Contextualised Representation
Entity Position-aware Sentence Representation
Bag Representation for Entity Pairs
NetAtt
NetMax
Collaborative Curriculum Learning
Objective Function
Curriculum Construction for Collaborative Learning
Dataset
Evaluation Metrics
Baseline Models
Effects of Internal CNNs Self-Attention
Effects of Collaborative Curriculum Learning
Acknowledgement b
Conclusion
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call