Multi-modal Contrastive Representation Learning for Entity Alignment

Proceedings of the 29th International Conference on Computational Linguistics, pages 2572–2584 October 12–17, 2022. 2572 Multi-modal Contrastive Representation Learning for Entity Alignment Zhenxi Lin1, Ziheng Zhang1, Meng Wang2, Yinghui Shi3, Xian Wu1, Yefeng Zheng1 1Tencent Jarvis Lab, Shenzhen, China 2School of Computer Science and Engineering, Southeast University, Nanjing, China 3School of Cyber Science and Engineering, Southeast University, Nanjing, China {chalerislin,zihengzhang}@tencent.com {meng.wang,shiyinghui}@seu.edu.cn {kevinxwu,yefengzheng}@tencent.com Abstract Multi-modal entity alignment aims to identify equivalent entities between two different multi- modal knowledge graphs, which consist of structural triples and images associated with entities. Most previous works focus on how to utilize and encode information from different modalities, while it is not trivial to leverage multi-modal knowledge in entity alignment be- cause of the modality heterogeneity. In this paper, we propose MCLEA, a Multi-modal Contrastive Learning based Entity Alignment model, to obtain effective joint representations for multi-modal entity alignment. Different from previous works, MCLEA considers task- oriented modality and models the inter-modal relationships for each entity representation. In particular, MCLEA firstly learns multiple in- dividual representations from multiple modali- ties, and then performs contrastive learning to jointly model intra-modal and inter-modal in- teractions. Extensive experimental results show that MCLEA outperforms state-of-the-art base- lines on public datasets under both supervised and unsupervised settings.1 1 Introduction Knowledge graphs (KGs) such as DBpe- dia (Lehmann et al., 2015) and YAGO (Mahdis- oltani et al., 2015) employ the graph structure to represent real-world knowledge, where the con- cepts are represented as nodes and the relationships among concepts are represented as edges. KGs have been widely applied to knowledge-driven applications to boost their performance, like recommendation system (Cao et al., 2019b), information extraction (Han et al., 2018) and question answering (Lan et al., 2021). In recent years, an increasing amount of knowledge has been represented in multi-modal formats, such as MMKG (Liu et al., 2019) and Richpedia (Wang 1The source code is available at https://github.com/lzxlin/ MCLEA. et al., 2020). These multi-modal KGs usually contain images as the visual modality, like profile photos, thumbnails, or posters. The augmented visual modality has shown the significant capability to improve KG-based applications (Chen et al., 2020a). It was also proven that the incorporation of visual modality can enhance the contextual semantics of entities and also achieve improved KG embeddings (Wang et al., 2021). Due to the large scope of real-world knowledge, most KGs are often incomplete, and multiple differ- ent KGs are usually complementary to one another. As a result, integrating multiple KGs into a unified one can enlarge the knowledge coverage and can also assist in refining KG by discovering the poten- tial flaws (Chen et al., 2020a). To integrate hetero- geneous multi-modal knowledge graphs, the task of multi-modal entity alignment (MMEA) is therefore proposed, which aims to discover equivalent enti- ties referring to the same real-world object. Several previous MMEA works have shown that the inclu- sion of visual modality in modeling helps to im- prove the performance of entity alignment. For in- stance, MMEA (Chen et al., 2020a) and EVA (Liu et …

Content truncated to 500 words