Subsequence-Level Entity Attention LSTM for Relation Extraction
Tao Gan, YUNQIANG GAN, Yanmin He · 2019
Relation extraction is a basic component in many knowledge base construction systems aiming at discovering structured facts from raw sentences. Traditional neural network models directly encode raw sentences to obtain semantic features based on which the relations are extracted. However, few efforts have been made to fully exploit the text between the entities that often contains important relation information. Therefore, our paper propose a subsequence-level entity attention LSTM network (EA-LSTM) for relation extraction. EA-LSTM first splits the input sentence into two subsequences which have duplicate content of the text between entities. An entity attention mechanism is then introduced in feature fusion stage to guide the model to selectively attend to the potential relation information of entities. The experimental results on the SemEval-2010 Task 8 dataset show that EA-LSTM is comparable to state-of-the-art systems even without using external features.