Joint Extraction Model of Entity and Relationship Based on Reinforcement Learning

In this paper, an entity relationship association extraction model based on reinforcement learning is proposed, namely, the sequence labeling Association model. It is used to solve the tag noise caused by the entity relationship extraction method based on remote monitoring. The model consists of two modules: sentence picker module and entity relation joint extraction module. First, the sentence picker module selects high-quality sentences with no data noise. Joint entity relationship extraction module will then selected sentences as its input and sequence annotation decoding method to forecast the input sentence and the sentence selected modules provide reward and guidance sentences selected sentences of choice high quality, and adopted a special label scheme combined extraction task is converted into a sequence labeling problem. Finally, the combination model of sequence annotation is trained to optimize sentence selection and sequence annotation together. The experimental results show that the sentence picker can effectively deal with the noise of the data, and the combination model of sequence annotation can effectively improve the F1 value of the sequence annotation model.