Multimodal Graph-Based Variational Mixture of Experts Network for Zero-Shot Multimodal Information Extraction

TL;DR

This work tackles zero-shot multimodal information extraction (ZS-MIE) by introducing MG-VMoE, a variational mixture-of-experts framework that aligns text and image representations at a fine-grained level using variational information bottlenecks. It couples a multimodal graph-based virtual adversarial training module to capture semantic correlations between samples, optimizing with a ranking objective and auxiliary losses to promote generalization to unseen categories. Empirical results on MET and MRE show MG-VMoE outperforms strong baselines, with notable gains in F1 and accuracy and clear ablation-supported contributions from the VMoE backbone and MG-VAT. The approach advances zero-shot capabilities in multimodal information extraction, enabling robust recognition of novel entity types and relations in social media data.

Abstract

Multimodal information extraction on social media is a series of fundamental tasks to construct the multimodal knowledge graph. The tasks aim to extract the structural information in free texts with the incorporate images, including: multimodal named entity typing and multimodal relation extraction. However, the growing number of multimodal data implies a growing category set and the newly emerged entity types or relations should be recognized without additional training. To address the aforementioned challenges, we focus on the zero-shot multimodal information extraction tasks which require using textual and visual modalities for recognizing unseen categories. Compared with text-based zero-shot information extraction models, the existing multimodal ones make the textual and visual modalities aligned directly and exploit various fusion strategies to improve their performances. But the existing methods ignore the fine-grained semantic correlation of text-image pairs and samples. Therefore, we propose the multimodal graph-based variational mixture of experts network (MG-VMoE) which takes the MoE network as the backbone and exploits it for aligning multimodal representations in a fine-grained way. Considering to learn informative representations of multimodal data, we design each expert network as a variational information bottleneck to process two modalities in a uni-backbone. Moreover, we also propose the multimodal graph-based virtual adversarial training to learn the semantic correlation between the samples. The experimental results on the two benchmark datasets demonstrate the superiority of MG-VMoE over the baselines.

Figures (6)

• Figure 1: The multimodal representation space comparison between the traditional MIE and the MG-VMoE models.
• Figure 2: The overall framework of the multimodal graph-based variational mixture of experts (MG-VMoE) network for zero-shot multimodal information extraction. The upper part is the samples of multimodal named entity typing and multimodal relation extraction. The lower left is the multimodal backbone network based on VMoE and the lower right is the multimodal graph-based vitrual adversarial training.
• Figure 3: The performance comparison of MG-VMoE with different expert module numbers.
• Figure 4: The performance comparison of MG-VMoE with different hyper-parameter $\beta$ values.
• Figure 5: The performance comparison of each expert module individually activated in MG-VMoE.