Relation Classification via CNN, Segmented Max-pooling, and SDP-BLSTM

Pengfei Wang, Zhipeng Xie, Junfeng Hu

Abstract: Relation classification is the task of classifying the semantic relation between two marked entities in a sentence. This paper proposes a novel neural model for this task. It first does convolution on input sentence to get local features of words in local context windows, and then designs a novel segmented max-pooling to reduce the temporal dimension from the length of sentence to the length of shortest dependency path (SDP) between two marked entities, and finally, a SDP-BLSTM network is applied to produce the final fixed-size vector representation of the relation instance, which is fed to a two-layer feed-forward network for classification. Experiments on the SemEval-2010 Task 8 dataset show that our model achieves competitive performance when compared with several start-of-the-art models.