Current hybrid approaches for LLM-KG fusion suffer from three core limitations: they introduce semantic noise during context augmentation (Ayoola et al., 2022; Xin et al., 2024), they remain constrained by LLM training biases in candidate generation (Ding Y. et al., 2024), and they create new modality-specific dependencies in multimodal fusion (Liu Q. et al., 2024).

Authors

• Person: Not available Organization: Frontiers
  Practices, opportunities and challenges in the fusion of knowledge ...

Sources

• Practices, opportunities and challenges in the fusion of knowledge ... www.frontiersin.org Frontiers via serper

Referenced by nodes (2)

• hybrid fusion approaches concept
• multimodal data integration concept