Multimodal Knowledge Learning for Named Entity Disambiguation

Zhang Dongjie, Longtao Huang


Abstract
With the popularity of online social media, massive-scale multimodal information has brought new challenges to traditional Named Entity Disambiguation (NED) tasks. Recently, Multimodal Named Entity Disambiguation (MNED) has been proposed to link ambiguous mentions with the textual and visual contexts to a predefined knowledge graph. Existing attempts usually perform MNED by annotating multimodal mentions and adding multimodal features to traditional NED models. However, these studies may suffer from 1) failing to model multimodal information at the knowledge level, and 2) lacking multimodal annotation data against the large-scale unlabeled corpus. In this paper, we explore a pioneer study on leveraging multimodal knowledge learning to address the MNED task. Specifically, we first harvest multimodal knowledge in the Meta-Learning way, which is much easier than collecting ambiguous mention corpus. Then we design a knowledge-guided transfer learning strategy to extract unified representation from different modalities. Finally, we propose an Interactive Multimodal Learning Network (IMN) to fully utilize the multimodal information on both the mention and knowledge sides. Extensive experiments conducted on two public MNED datasets demonstrate that the proposed method achieves improvements over the state-of-the-art multimodal methods.
Anthology ID:
2022.findings-emnlp.230
Volume:
Findings of the Association for Computational Linguistics: EMNLP 2022
Month:
December
Year:
2022
Address:
Abu Dhabi, United Arab Emirates
Editors:
Yoav Goldberg, Zornitsa Kozareva, Yue Zhang
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
3160–3169
Language:
URL:
https://aclanthology.org/2022.findings-emnlp.230
DOI:
10.18653/v1/2022.findings-emnlp.230
Bibkey:
Cite (ACL):
Zhang Dongjie and Longtao Huang. 2022. Multimodal Knowledge Learning for Named Entity Disambiguation. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 3160–3169, Abu Dhabi, United Arab Emirates. Association for Computational Linguistics.
Cite (Informal):
Multimodal Knowledge Learning for Named Entity Disambiguation (Dongjie & Huang, Findings 2022)
Copy Citation:
PDF:
https://aclanthology.org/2022.findings-emnlp.230.pdf