.png){kind=icon}
.png){kind=icon}
Return
Dynamic cognitive cycle-driven multimodal agent for knowledge graph completion
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2026.132001.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• First Agent-Based Paradigm for Multi-Modal Knowledge Graph Completion (MMKGC): We propose an MMKGC-Agent framework that reformulates multimodal knowledge graph completion from static embedding learning into a context-aware, closed-loop dynamic reasoning paradigm. • We design a Learnable Residual Memory Pathway (LRMP) to actively construct query-dependent local memory states for adaptive inference. • We introduce a Semantic Self-Alignment mechanism via Barlow Twins to enforce intrinsic representational coherence and purify multimodal noise. • Extensive experiments on multiple MMKGC benchmarks show consistent gains, achieving state-of-the-art results on four datasets (DB15K, MKG-W, MKG-Y, KVC16K), especially in top-1 accuracy.
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W