Complete multi-modal metric learning for multi-modal sarcasm detection

Jiecheng Zhang, C.L.Philip Chen, Shuzhen Li, Tong Zhang

Keywords: multi-modal sarcasm detection, metric learning, complete multi-modal incongruities

TL;DR: CMML-Net is the first work in multi-modal sarcasm detection to introduce deep metric learning to explicitly capture complete multi-modal incongruities in fact and sentiment perspectives.

Abstract:

Multi-modal sarcasm detection identifies sarcasm from text-image pairs, an essential technology for accurately understanding the user's real attitude.
Most research extracted the incongruity of text-image pairs as sarcasm information. However, these methods neglected inter-modal or intra-modal incongruities in fact and sentiment perspectives, leading to incomplete sarcasm information and biased performance.
To address the above issues, this paper proposes a complete multi-modal metric learning network (CMML-Net) for multi-modal sarcasm detection tasks.
Specifically, CMML-Net utilizes a fact-sentiment multi-task representation learning module to produce refined fact and sentiment text-image representation pairs.
It then designs a complete multi-modal metric learning to iteratively calculate inter-modal and intra-modal incongruities in a unified space (e.g., fact and sentiment metric space), efficiently capturing complete multi-modal incongruities.
CMML-Net performs well in explicitly capturing comprehensive sarcasm information and obtaining discriminative performance via deep metric learning.
The state-of-the-art performance on the widely-used dataset demonstrates CMML-Net's effectiveness in multi-modal sarcasm detection.

Primary Area: applications to computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 13962