Distant Supervision for Entity Relation Extraction

Aiming at the extreme shortage of high-quality relationship extraction data sets in the field of network security and the small scale of self-labeled data sets, a piecewise convolutional neural network model (DoubleENT-Rel-PCNN, DRPCNN) is proposed to enhance double-end entity information and relationship information. The multi-kernel convolution layer performs multi-dimensional convolution operation on the input text feature vector to capture the hidden layer representation of the text from multiple perspectives. The weight of each sentence in the example bag is calculated. Finally, the confidence between each relation category and the example bag feature vector is calculated, and the relation label of the example bag is predicted according to the confidence. The experimental results show that compared with the PCNN-Attention baseline model, the AUC value is increased by $3.6 \%$, which demonstrates the effectiveness of the proposed piecewise convolutional neural network model for entity relation extraction with enhanced double-end entity information and relationship information.