近日，哈工大社会计算与信息检索研究中心的一支研究团队取得了重要突破，提出了一个创新的迭代推理框架DPE-MNER。该框架致力于推进多模态命名实体识别技术的发展，作为构建多模态知识图谱的基础关键任务。
研究团队指出，多模态命名实体识别要求研究者整合多种模态信息，以精准地从文本中提取命名实体。尽管过去的研究已经探索了多模态表示的整合方法，但在融合多模态表示以提供丰富上下文信息、提升多模态命名实体识别的性能方面仍有不足。
针对这一问题，DPE-MNER框架应运而生。研究团队采用“分解、优先、消除”的策略，动态整合多样化的多模态表示。该框架将多模态表示的融合巧妙地分解为层次化且相互连接的融合层，从而极大地简化了处理过程。这一创新性的设计使得框架能够在复杂的多模态信息中更加精准地识别和提取实体。
哈工大的这支研究团队由郑子豪、张梓涵、王泽鑫、付瑞吉、刘铭、王仲远和秦兵等人构成，他们的成果已经在AIxiv专栏上发布，该栏目是机器之心发布学术、技术内容的栏目，有效促进了学术交流与传播。
此研究成果的发布，标志着我国在多模态知识图谱构建领域取得了重要进展。该框架的提出，将对相关领域的研究产生深远影响，并有望为未来的知识图谱构建提供新的思路和方法。
英语如下：
News Title: Harbin Institute of Technology’s Innovative Iterative Reasoning Framework DPE-MNER Ushers in a New Breakthrough in Multi-modal Named Entity Recognition
Keywords: 1. Harbin Institute of Technology proposes DPE-MNER innovative framework
News Content:
Harbin Institute of Technology research team proposes an innovative iterative reasoning framework, DPE-MNER, to promote the development of multi-modal named entity recognition technology
Recently, a research team from the Social Computing and Information Retrieval Research Center at Harbin Institute of Technology has made significant breakthroughs and proposed an innovative iterative reasoning framework, DPE-MNER. This framework is dedicated to advancing the development of multi-modal named entity recognition technology as a fundamental task for building a multi-modal knowledge graph.
The research team pointed out that multi-modal named entity recognition requires researchers to integrate multiple modal information to accurately extract named entities from text. Although past studies have explored methods of integrating multi-modal representations, there are still deficiencies in providing rich contextual information and enhancing the performance of multi-modal named entity recognition.
To address this issue, the DPE-MNER framework has been developed. The research team uses a “decomposition, prioritization, elimination” strategy to dynamically integrate diverse multi-modal representations. This framework cleverly decomposes the integration of multi-modal representations into hierarchical and interconnected fusion layers, greatly simplifying the processing process. This innovative design enables the framework to more accurately identify and extract entities from complex multi-modal information.
The Harbin Institute of Technology research team consists of Zheng Zihao, Zhang Zihan, Wang Zexin, Fu Ruiji, Liu Ming, Wang Zhongyuan, and Qin Bing. Their achievements have been published in the AIxiv column, which is a column for publishing academic and technical content by Machine Mind, effectively promoting academic exchanges and dissemination.
The publication of this research achievement marks significant progress in the field of multi-modal knowledge graph construction in China. The proposal of this framework will have a far-reaching impact on related fields and is expected to provide new ideas and methods for future knowledge graph construction.
【来源】https://www.jiqizhixin.com/articles/2024-07-02-3
Views: 1