Neural Relation Classification with Text Descriptions

Proceedings of the 27th International Conference on Computational Linguistics, pages 1167–1177 Santa Fe, New Mexico, USA, August 20-26, 2018. 1167 Neural Relation Classification with Text Descriptions Feiliang Ren, Di Zhou, Zhihui Liu, Yongcheng Li, Rongsheng Zhao, Yongkang Liu, Xiaobo Liang School of Computer Science and Engineering, Northeastern University, Shenyang, 110819, China renfeiliang@cse.neu.edu.cn Abstract Relation classification is an important task in natural language processing fields. State-of-the-art methods usually concentrate on building deep neural networks based classification models on the training data in which the relations of the labeled entity pairs are given. However, these methods usually suffer from the data sparsity issue greatly. On the other hand, we notice that it is very easy to obtain some concise text descriptions for almost all of the entities in a relation classification task. The text descriptions can provide helpful supplementary information for relation classifi- cation. But they are ignored by most of existing methods. In this paper, we propose DesRC, a new neural relation classification method which integrates entities text descriptions into deep neural networks models. We design a two-level attention mechanism to select the most useful information from the ”intra-sentence” aspect and the ”cross-sentence” aspect. Besides, the ad- versarial training method is also used to further improve the classification performance. Finally, we evaluate the proposed method on the SemEval 2010 dataset. Extensive experiments show that our method achieves much better experimental results than other state-of-the-art relation classification methods. 1 Introduction The aim of relation classification is that given a sentence in which two target entities are labeled, to select a proper relation for these two entities from a predefined relation set. For example, given a sen- tence “The system as described above has its greatest application in an arrayed <e1>configuration </e1> of antenna <e2>elements</e2>”, a relation classification system aims to identify that there is a “Component-Whole” relation from e2 to e1. Obviously, accurate relation classification results would benefit lots of natural language processing tasks, such as sentence interpretations, Q&A, knowledge graph construction, ontology learning, and so on. Thus, lots of researchers have devoted to this research field. For relation classification, deep neural networks (DNN) based methods have been widely explored and have achieved state-of-the-art experimental results. However, when evaluating state-of-the-art relation classification methods on some standard datasets, experimental results show that there are usually huge performance gaps between different relations. Besides, when using the trained relation classification models to predict relations on new data, the prediction performance is usually far lower than expected. This is mainly because of the data sparsity issue: first, there are always some relations that have far less training data than others; second, the available training data is not sufficient enough to train a robust relation classification model. On the other hand, we notice that for almost all of the entities in a relation classification task, there are usually available text descriptions for them on some Encyclopedia websites like Wikipedia, DBpedia, Wikidata, etc. For example, from the Wikipedia website, we can extract the following text descriptions as shown …

Content truncated to 500 words