多任务学习和互信息最大化与跨模态变换器用于多模态情感分析

IF 2.3 3区 计算机科学 Q3 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE
Yang Shi, Jinglang Cai, Lei Liao
{"title":"多任务学习和互信息最大化与跨模态变换器用于多模态情感分析","authors":"Yang Shi, Jinglang Cai, Lei Liao","doi":"10.1007/s10844-024-00858-9","DOIUrl":null,"url":null,"abstract":"<p>The effectiveness of multimodal sentiment analysis hinges on the seamless integration of information from diverse modalities, where the quality of modality fusion directly influences sentiment analysis accuracy. Prior methods often rely on intricate fusion strategies, elevating computational costs and potentially yielding inaccurate multimodal representations due to distribution gaps and information redundancy across heterogeneous modalities. This paper centers on the backpropagation of loss and introduces a Transformer-based model called Multi-Task Learning and Mutual Information Maximization with Crossmodal Transformer (MMMT). Addressing the issue of inaccurate multimodal representation for MSA, MMMT effectively combines mutual information maximization with crossmodal Transformer to convey more modality-invariant information to multimodal representation, fully exploring modal commonalities. Notably, it utilizes multi-modal labels for uni-modal training, presenting a fresh perspective on multi-task learning in MSA. Comparative experiments on the CMU-MOSI and CMU-MOSEI datasets demonstrate that MMMT improves model accuracy while reducing computational burden, making it suitable for resource-constrained and real-time performance-requiring application scenarios. Additionally, ablation experiments validate the efficacy of multi-task learning and probe the specific impact of combining mutual information maximization with Transformer in MSA.</p>","PeriodicalId":56119,"journal":{"name":"Journal of Intelligent Information Systems","volume":"16 1","pages":""},"PeriodicalIF":2.3000,"publicationDate":"2024-07-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Multi-task learning and mutual information maximization with crossmodal transformer for multimodal sentiment analysis\",\"authors\":\"Yang Shi, Jinglang Cai, Lei Liao\",\"doi\":\"10.1007/s10844-024-00858-9\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p>The effectiveness of multimodal sentiment analysis hinges on the seamless integration of information from diverse modalities, where the quality of modality fusion directly influences sentiment analysis accuracy. Prior methods often rely on intricate fusion strategies, elevating computational costs and potentially yielding inaccurate multimodal representations due to distribution gaps and information redundancy across heterogeneous modalities. This paper centers on the backpropagation of loss and introduces a Transformer-based model called Multi-Task Learning and Mutual Information Maximization with Crossmodal Transformer (MMMT). Addressing the issue of inaccurate multimodal representation for MSA, MMMT effectively combines mutual information maximization with crossmodal Transformer to convey more modality-invariant information to multimodal representation, fully exploring modal commonalities. Notably, it utilizes multi-modal labels for uni-modal training, presenting a fresh perspective on multi-task learning in MSA. Comparative experiments on the CMU-MOSI and CMU-MOSEI datasets demonstrate that MMMT improves model accuracy while reducing computational burden, making it suitable for resource-constrained and real-time performance-requiring application scenarios. Additionally, ablation experiments validate the efficacy of multi-task learning and probe the specific impact of combining mutual information maximization with Transformer in MSA.</p>\",\"PeriodicalId\":56119,\"journal\":{\"name\":\"Journal of Intelligent Information Systems\",\"volume\":\"16 1\",\"pages\":\"\"},\"PeriodicalIF\":2.3000,\"publicationDate\":\"2024-07-10\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Journal of Intelligent Information Systems\",\"FirstCategoryId\":\"94\",\"ListUrlMain\":\"https://doi.org/10.1007/s10844-024-00858-9\",\"RegionNum\":3,\"RegionCategory\":\"计算机科学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q3\",\"JCRName\":\"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Intelligent Information Systems","FirstCategoryId":"94","ListUrlMain":"https://doi.org/10.1007/s10844-024-00858-9","RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE","Score":null,"Total":0}
引用次数: 0

摘要

多模态情感分析的有效性取决于对不同模态信息的无缝整合,而模态融合的质量直接影响情感分析的准确性。先前的方法通常依赖于复杂的融合策略,从而提高了计算成本,并且由于异构模态之间的分布差距和信息冗余,可能会产生不准确的多模态表示。本文以损失的反向传播为中心,介绍了一种基于变换器的模型,称为跨模态变换器的多任务学习和互信息最大化(MMMT)。针对 MSA 多模态表征不准确的问题,MMMT 将互信息最大化与跨模态变换器有效结合,为多模态表征传递更多模态不变信息,充分挖掘模态共性。值得注意的是,它利用多模态标签进行单模态训练,为 MSA 的多任务学习提供了一个全新的视角。在 CMU-MOSI 和 CMU-MOSEI 数据集上进行的对比实验表明,MMMT 提高了模型的准确性,同时减轻了计算负担,使其适用于资源受限和要求实时性能的应用场景。此外,消融实验验证了多任务学习的功效,并探究了在 MSA 中将互信息最大化与 Transformer 相结合的具体影响。
本文章由计算机程序翻译,如有差异,请以英文原文为准。

Multi-task learning and mutual information maximization with crossmodal transformer for multimodal sentiment analysis

Multi-task learning and mutual information maximization with crossmodal transformer for multimodal sentiment analysis

The effectiveness of multimodal sentiment analysis hinges on the seamless integration of information from diverse modalities, where the quality of modality fusion directly influences sentiment analysis accuracy. Prior methods often rely on intricate fusion strategies, elevating computational costs and potentially yielding inaccurate multimodal representations due to distribution gaps and information redundancy across heterogeneous modalities. This paper centers on the backpropagation of loss and introduces a Transformer-based model called Multi-Task Learning and Mutual Information Maximization with Crossmodal Transformer (MMMT). Addressing the issue of inaccurate multimodal representation for MSA, MMMT effectively combines mutual information maximization with crossmodal Transformer to convey more modality-invariant information to multimodal representation, fully exploring modal commonalities. Notably, it utilizes multi-modal labels for uni-modal training, presenting a fresh perspective on multi-task learning in MSA. Comparative experiments on the CMU-MOSI and CMU-MOSEI datasets demonstrate that MMMT improves model accuracy while reducing computational burden, making it suitable for resource-constrained and real-time performance-requiring application scenarios. Additionally, ablation experiments validate the efficacy of multi-task learning and probe the specific impact of combining mutual information maximization with Transformer in MSA.

求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
Journal of Intelligent Information Systems
Journal of Intelligent Information Systems 工程技术-计算机:人工智能
CiteScore
7.20
自引率
11.80%
发文量
72
审稿时长
6-12 weeks
期刊介绍: The mission of the Journal of Intelligent Information Systems: Integrating Artifical Intelligence and Database Technologies is to foster and present research and development results focused on the integration of artificial intelligence and database technologies to create next generation information systems - Intelligent Information Systems. These new information systems embody knowledge that allows them to exhibit intelligent behavior, cooperate with users and other systems in problem solving, discovery, access, retrieval and manipulation of a wide variety of multimedia data and knowledge, and reason under uncertainty. Increasingly, knowledge-directed inference processes are being used to: discover knowledge from large data collections, provide cooperative support to users in complex query formulation and refinement, access, retrieve, store and manage large collections of multimedia data and knowledge, integrate information from multiple heterogeneous data and knowledge sources, and reason about information under uncertain conditions. Multimedia and hypermedia information systems now operate on a global scale over the Internet, and new tools and techniques are needed to manage these dynamic and evolving information spaces. The Journal of Intelligent Information Systems provides a forum wherein academics, researchers and practitioners may publish high-quality, original and state-of-the-art papers describing theoretical aspects, systems architectures, analysis and design tools and techniques, and implementation experiences in intelligent information systems. The categories of papers published by JIIS include: research papers, invited papters, meetings, workshop and conference annoucements and reports, survey and tutorial articles, and book reviews. Short articles describing open problems or their solutions are also welcome.
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信