.png){kind=icon}
Return
World Knowledge Enhanced Task-Oriented Semantic Communication Based on Large Language Model
.png){kind=icon}
PRE
AI
DOI：10.1109/tvt.2026.3737391.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Semantic communication (SC) has emerged as a promising technology in next generation wireless networks for its energy-efficient, high-performance, and intelligent task execution capabilities. Recent studies have utilized the powerful semantic understanding and representation abilities of large language models (LLMs) to enhance the performance of SC. However, most existing works overlook the diverse and complex physical environments that the edge devices encounter while performing intelligent tasks, which can significantly hinder their task execution effectiveness in real-world scenarios. To address this challenge, we propose a world knowledge enhanced task-oriented SC based on LLM, which can improve the downstream physical world reasoning capability of the receiver side LLM without compromising the generality. Specifically, a knowledge extraction method is developed to acquire diverse embodied experiences from the world model. Besides, a multi-task regularized finetuning framework is designed to improve training speed and enable efficient continual learning of LLM. Simulation results verify that the proposed scheme enhances the spatial reasoning abilities of the SC system in task execution and outperforms other baseline methods across various channels.
Keywords:
Semantic communication
world knowledge
large language model
real-world scenario
Journal
IF:
7.1
Papers:
1.8W
Citations:
6.6W
Organization
Cited Papers
No cited papers available