Silk Road Multilingual Cognitive Computing International Cooperation Joint Laboratory, School of Computer Science and Technology (School of Cyberspace Security), Xinjiang University, Urumqi, Xinjiang, China
Auto-generated profile, based on publicly available information
Jianshe Wang
Silk Road Multilingual Cognitive Computing International Cooperation Joint Laboratory, School of Computer Science and Technology (School of Cyberspace Security), Xinjiang University, Urumqi, Xinjiang, China
Uncertain Knowledge Graph Reasoning (UKGR) aims to infer missing facts and their confidence levels from incomplete and uncertain knowledge graphs. However, existing methods face dual challenges: traditional methods struggle to model uncertain information, while existing large language model (LLM)-based approaches fail to effectively utilize the critical uncertainty-structured information of knowledge graphs during reasoning. To address this core issue, we propose a Topology-Driven Prefix Adapter (TDPA), which activates the reasoning potential of LLMs by deeply integrating confidence topology features of uncertain knowledge graphs. In implementation, TDPA uses graph embedding to comprehend the intricate entities and relations within KGs, representing them as structural embeddings. Then TDPA communicates such structural information understanding to the LLMs through a knowledge prefix adapter which projects the structural embeddings into the textual space and obtains virtual knowledge tokens positioned as a prefix of the input prompt. Experiments on CN15K, NL27K, and PPI5K datasets demonstrate that our method outperforms traditional probabilistic graph models and existing LLM-based approaches on MSE and MAE metrics, confirming that effectively modeling uncertainty topology structures is key to enhancing the reasoning capabilities of LLMs.