Learning mention and relation representation with convolutional neural networks for relation extraction

Most previous works for relation extraction are based on handcrafting sentence-level features such as part of speech, named entity and dependency tree path properties. This paper proposes a new approach to learn the embedding of the mentions and their relations using convolutional neural networks with a pairwise ranking loss function. Through the learned mention and relation embeddings we can get a score to evaluate the relevance of a given pair of mention and relation. We show that our approach using word embeddings as input features for our model can learn better mention and relation representation and is superior to state-of-the-art results.