Reinforcement Learning for Relation Classification from Noisy Data
This work addresses the noisy labeling problem in relation classification for natural language processing, offering an incremental improvement over existing distant supervision methods.
The paper tackles the problem of relation classification from noisy data by proposing a novel model with an instance selector and a relation classifier trained jointly using reinforcement learning, resulting in effective noise handling and improved performance at the sentence level.
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.