ActiveEA: Active Learning for Neural Entity Alignment

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 3364–3374 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 3364 ActiveEA: Active Learning for Neural Entity Alignment Bing Liu1, �, Harrisen Scells1, Guido Zuccon1, Wen Hua1, Genghong Zhao2 1The University of Queensland, Australia 2Neusoft Research of Intelligent Healthcare Technology, Co. Ltd., China {bing.liu, h.scells, g.zuccon, w.hua}@uq.edu.au zhaogenghong@neusoft.com Abstract Entity Alignment (EA) aims to match equival- ent entities across different Knowledge Graphs (KGs) and is an essential step of KG fu- sion. Current mainstream methods – neural EA models – rely on training with seed align- ment, i.e., a set of pre-aligned entity pairs which are very costly to annotate. In this paper, we devise a novel Active Learning (AL) frame- work for neural EA, aiming to create highly informative seed alignment to obtain more ef- fective EA models with less annotation cost. Our framework tackles two main challenges encountered when applying AL to EA: (1) How to exploit dependencies between entit- ies within the AL strategy. Most AL strategies assume that the data instances to sample are independent and identically distributed. How- ever, entities in KGs are related. To address this challenge, we propose a structure-aware uncertainty sampling strategy that can measure the uncertainty of each entity as well as its im- pact on its neighbour entities in the KG. (2) How to recognise entities that appear in one KG but not in the other KG (i.e., bachel- ors). Identifying bachelors would likely save annotation budget. To address this challenge, we devise a bachelor recognizer paying atten- tion to alleviate the effect of sampling bias. Empirical results show that our proposed AL strategy can significantly improve sampling quality with good generality across different datasets, EA models and amount of bachelors. 1 Introduction Knowledge Graphs (KGs) store entities and their relationships with a graph structure and are used as knowledge drivers in many applications (Ji et al., 2020). Existing KGs are often incomplete but com- plementary to each other. A popular approach used to tackle this problem is KG fusion, which attempts to combine several KGs into a single, comprehens- ive one. Entity Alignment (EA) is an essential Republican Party Donald Trump New York City 14 June, 1946 US BirthDate BornIn MemberOf PresidentOf Trump Tower D.J. Trump The Trump Organization America Wealth OwnerOf OwnerOf PresidentOf 2.5 Billion USD Political knowledge Business knowledge Figure 1: An example of Entity Alignment. step for KG fusion: it identifies equivalent entities across different KGs, supporting the unification of their complementary knowledge. For example, in Fig. 1 Donald Trump and US in the first KG cor- respond to D.J. Trump and America respectively in the second KG. By aligning them, the political and business knowledge about Donald Trump can be integrated within one KG. Neural models (Chen et al., 2017, 2018; Wang et al., 2018; Cao et al., 2019) are the current state- of-the-art in EA and are capable of matching en- tities in an end-to-end manner. Typically, these neural EA models rely …

Content truncated to 500 words