.png){kind=icon}
.png){kind=icon}
Return
AlignExperts: Knowledge alignment experts for summarization faithfulness evaluation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2025.130943.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• Two-stage model uses knowledge-aligned experts for faithfulness evaluation. • Alignment experts enable solo or joint evaluation, cutting performance trade-offs. • Achieves benchmark gains without error annotations or augmentation. • Surpasses LLM-based faithfulness models in performance. • Aligning knowledge first then fine-tuning reduces parameters.
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W