EClaiRE: Context Matters! – Comparing Word Embeddings for Relation Classification

Abstract

In recent years, there has been an increasing interest in the task of relation classification, which aims to label a relation between two semantic entities. In this work, we investigate how domain-specific information influences the performance of ClaiRE, an SVM-based system combining manually crafted features with word embeddings. To this end, we experiment with a wide range of word embeddings and evaluate on one general and two scientific relation classification datasets. We release all of our code for relation classification and data for scientific word embeddings to enable the reproduction of our experiments.