Classifying Relations by Ranking with Convolutional Neural Networks

IJCNLP 2015  ·  Cicero Nogueira dos Santos, Bing Xiang, Bo-Wen Zhou ·

Relation classification is an important semantic processing task for which state-ofthe-art systems still rely on costly handcrafted features. In this work we tackle the relation classification task using a convolutional neural network that performs classification by ranking (CR-CNN). We propose a new pairwise ranking loss function that makes it easy to reduce the impact of artificial classes. We perform experiments using the the SemEval-2010 Task 8 dataset, which is designed for the task of classifying the relationship between two nominals marked in a sentence. Using CRCNN, we outperform the state-of-the-art for this dataset and achieve a F1 of 84.1 without using any costly handcrafted features. Additionally, our experimental results show that: (1) our approach is more effective than CNN followed by a softmax classifier; (2) omitting the representation of the artificial class Other improves both precision and recall; and (3) using only word embeddings as input features is enough to achieve state-of-the-art results if we consider only the text between the two target nominals.

PDF Abstract IJCNLP 2015 PDF IJCNLP 2015 Abstract

Datasets


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Relation Extraction SemEval-2010 Task-8 CR-CNN F1 84.1 # 27

Methods