MM-TComplEx:基于时间感知的多模态时序知识图谱嵌入框架
陈嘉豪, 胡锦宗, 龙怡梅, 田欣, 钱程, 陈雅茜
西南民族大学学报(自然科学版) ›› 2026, Vol. 52 ›› Issue (4) : 428 -434.
MM-TComplEx:基于时间感知的多模态时序知识图谱嵌入框架
MM-TComplEx: a time-aware-based multi-modal temporal knowledge graph embedding framework
传统静态知识图谱难以建模时序动态性与多模态信息,限制了推理、问答等复杂任务的应用;当前多模态时序知识图谱嵌入方法还存在单一模态局限、隐含时间信息处理准确率低等问题.为此,提出时间感知约束的MM-TComplEx框架:通过自适应门控融合与多专家网络实现多模态对齐,扩展TComplEx至复数空间以融合时序与多模态信息,并采用混合损失函数保障语义及时序一致性.实验显示,在三个基准数据集上,MM-TComplEx 的实体预测与时间预测指标,均显著优于TComplEx、MPlexGNN等基线模型.
It is difficult for traditional static knowledge graphs to model temporal dynamics and multimodal information, which limits their application in complex tasks such as reasoning and question answering. The existing methods for multimodal temporal knowledge graph embedding are still faced with limitations such as single modality limitations and low accuracy in handling implicit temporal information. To address these issues, MM-TComplEx, a time-aware constrained framework, was proposed. Specifically, multimodal alignment was achieved through adaptive gated fusion and multi-expert network, TComplEx was further extended into the complex space to integrate temporal and multimodal information, and a hybrid loss function was adopted to ensure both semantic and temporal consistency. Experimental results showed that MM-TComplEx significantly outperformed baseline models such as TComplEx and MPlexGNN in both entity prediction and time prediction tasks on three benchmark datasets.
| [1] |
葛依琳, 孙海春, 袁得嵛. 融合多模态知识与有监督检索的视觉问答模型[J]. 计算机科学与探索, 2025,19(8): 2203-2218. |
| [2] |
吴宗航, 张东, 李冠宇. 基于联合自监督学习的多模态融合推荐算法[J]. 计算机应用, 2025,45(6):1858-1868. |
| [3] |
张换香, 李梦云, 张景. 基于多模态信息融合的中文隐式情感分析[J]. 计算机工程与应用, 2025,61(2):179-190. |
| [4] |
范勇, 张政波, 王晶. 基于深度学习的电子病历多模态数据融合研究进展[J]. 生物医学工程学杂志, 2024,41(5): 1062-1071. |
| [5] |
马浩然, 王金华. 基于图结构与元素信息融合的实体对齐技术[J]. 科技导报, 2024,42(18):98-109. |
| [6] |
陈囿任, 李勇, 温明, |
| [7] |
徐金诚, 葛云生. TransE-KCB: 一种改进负样本采样的知识图谱表示方法[J]. 计算机应用与软件, 2024,41(8):345-350. |
| [8] |
李忠良, 陈麒, 石琳, |
| [9] |
王雪钰. 基于知识推理的图谱补全算法研究与实现[D]. 北京: 北京邮电大学, 2024. |
| [10] |
|
| [11] |
|
| [12] |
陈强, 张栋, 李寿山, |
| [13] |
刘畅成, 桑磊, 李炜, |
| [14] |
朱玉亮, 刘俊涛, 饶子昀, |
| [15] |
武月佳, 周建涛. DL~+:一种增强型双层知识图谱推理框架[J]. 计算机科学, 2023,50(12):302-313. |
| [16] |
陈博洋. 基于规则的知识图谱推理算法研究与实现[D]. 北京: 北京邮电大学, 2023. |
| [17] |
|
| [18] |
|
| [19] |
|
| [20] |
|
/
| 〈 |
|
〉 |