Word Embedding Representation with Synthetic Position and Context Information for Relation Extraction

Abstract

In recent years, various knowledge bases have been built and widely used in different natural language possessing tasks. And relation extraction is an effective way to enrich knowledge bases. But in most existing relation extraction methods, they obtain word embedding from pre-trained Word2vec or GloVe, which don't consider the difference of word in different sentences. But, such a fact cannot be ignored, that is, the same word in different contexts or in different position in a sentence has different meanings. So, we propose an approach to get word embedding representation with synthetic context and position information and call it semantic word embedding. After getting semantic word embedding, we can get sentence-level representation by simple average-pooling rather than complex architecture of convolutional neural network. Furthermore, we apply the semantic word embedding representation to the relation extraction task of Natural Language Processing. The experimental results show that the performance of the proposed method on the popular benchmark dataset is better than the state-of-the-art CNN-based approach.