Reinforcement Learning for Relation Classification from Noisy Data

TL;DR

This paper introduces a novel sentence-level relation classification model that effectively handles noisy data by jointly training an instance selector and a relation classifier using reinforcement learning, improving accuracy over existing bag-level methods.

Contribution

The paper presents a new model with a reinforcement learning-based instance selector for sentence-level relation classification from noisy data, addressing limitations of previous bag-level approaches.

Findings

Improved relation classification accuracy on noisy datasets

Effective noise handling through joint training of modules

Outperforms existing bag-level methods

Abstract

Existing relation classification methods that rely on distant supervision assume that a bag of sentences mentioning an entity pair are all describing a relation for the entity pair. Such methods, performing classification at the bag level, cannot identify the mapping between a relation and a sentence, and largely suffers from the noisy labeling problem. In this paper, we propose a novel model for relation classification at the sentence level from noisy data. The model has two modules: an instance selector and a relation classifier. The instance selector chooses high-quality sentences with reinforcement learning and feeds the selected sentences into the relation classifier, and the relation classifier makes sentence level prediction and provides rewards to the instance selector. The two modules are trained jointly to optimize the instance selection and relation classification processes. Experiment results show that our model can deal with the noise of data effectively and obtains better performance for relation classification at the sentence level.