A bootstrapping and MV-RNN mixed method for relation extraction

The classical method for information extraction called bootstrapping is widely used for its good performance. But some inevitable weakness such as semantic drift and low recall hinders the improvement of the model. With the repopulation of neural network, some neural network methods also showed their power in relation extraction with the loss of accuracy compared to bootstrapping method. Under this circumstance, we propose a method, which mixed these two methods together and combined their pros together.