The Combination of CNN, RNN, and DNN for Relation Extraction

Abstract

Relation extraction, which is a subtask of NLP (natural language processing) field, its target is to identify the entities in texts and extract the relation between entities. Previous works prove that neural networks are feasible for relation extraction. CNN (convolutional neural networks) and LSTM (long short-term memory) are two majority models used in relation extraction. Further research shows that the combination of CNN and LSTM has a better performance. Inspired by the solution of LVCSR (Large-Vocabulary-Continuous-Speech-Recognition), another task in the NLP field, we propose adding DNN after the combination of CNN and LSTM. This model achieves a better effect on the precision-recall curve than the previous model.