Data Augmentation for Few-Shot Knowledge Graph Completion from Hierarchical Perspective

Proceedings of the 29th International Conference on Computational Linguistics, pages 2494–2503 October 12–17, 2022. 2494 Data Augmentation for Few-Shot Knowledge Graph Completion from Hierarchical Perspective Yuanzhou Yao 1,2, Zhao Zhang 1,3∗, Yongjun Xu 1 and Chao Li 3,1∗ 1 Institute of Computeing Technology, Chinese Academy of Sciences, Beijing, China 2 University of Chinese Academy of Sciences, Beijing, China 3 Zhejiang Lab, Hangzhou, China {yaoyuanzhou21s, zhangzhao2021, xyj, lichao}@ict.ac.cn Abstract Few-shot knowledge graph completion (FKGC) has become a new research focus in the field of knowledge graphs in recent years, which aims to predict the missing links for re- lations that only have a few associative triples. Existing models attempt to solve the problem via learning entity and relation representations. However, the limited training data severely hinders the performance of existing models. To this end, we propose to solve the FKGC problem with the data augmentation technique. Specifically, we perform data augmentation from two perspectives, i.e., inter-task view and intra-task view. The former generates new tasks for FKGC, while the latter enriches the support or query set for an individual task. It is worth noting that the proposed framework can be applied to a number of existing FKGC models. Experimental evaluation on two public datasets indicates our model is capable of achieving substantial improvements over baselines. 1 Introduction Knowledge graphs (KGs) are structured seman- tic knowledge bases used to describe concepts and their interrelationships in the physical world in symbolic form. Many KGs in the real world, such as Freebase (Bollacker et al., 2008), YAGO (Suchanek et al., 2007), WordNet (Miller, 1992), Wikidata (Vrandecic and Krötzsch, 2014) and NELL (Mitchell et al., 2018), consist of triple facts in the form of (head entity, relation, tail entity), e.g., (Paris, capitalOf, France) indicates that Paris is the capital of France. KGs have been introduced into various downstream tasks of NLP, such as ques- tion answering (Saxena et al., 2020), dialogue sys- tems (He et al., 2017) and information extraction (Hoffmann et al., 2011), etc. The integrity of KG promotes the performance of downstream tasks. ∗ Corresponding author: Zhao Zhang and Chao Li. However, KGs in the real world are far from com- plete and comprehensive. Therefore, it is necessary to complete KGs by inferring new triple facts. To complete KGs, most existing embedding- based KG completion models require adequate triples for each relation as training data, such as TransE (Bordes et al., 2013), RotatE (Sun et al., 2019) and ConvE (Dettmers et al., 2017). However, in reality, the number of triples for each relation conforms to a long-tail distribution (Xiong et al., 2018), i.e., only a small number of relations oc- cur frequently, while most relations only occur a few times in a KG. This phenomenon hinders to learning reliable representations for infrequent re- lations and further degrades the KG completion performance. This has motivated an emerging research topic named few-shot knowledge graph completion (FKGC), where one task is to predict the tail en- tity t in a query (h, r, ?) given only …

Content truncated to 500 words