Institution profile

Beijing Key Laboratory of Big Data Management and Analysis Methods

Academic institutionasia · cn
Research library1linked papers
Opportunities0open roles
Selected work

Representative Papers

Tackling the Inherent Difficulty of Noise Filtering in RAG

Jan 05, 2026arXiv.org

This work addresses the vulnerability of retrieval-augmented generation (RAG) systems to performance degradation and hallucination caused by irrelevant or noisy retrieved documents. To overcome the limitations of existing approaches in effectively filtering such noise, the paper proposes a novel fine-tuning strategy that transcends conventional constraints on attention architecture. Specifically, the method introduces a tailored training objective and targeted modifications to the attention mechanism to enhance the model’s ability to discriminate between relevant and irrelevant retrieved content. This approach substantially improves the model’s capacity for information filtering and robustness in noisy retrieval settings. Experimental results demonstrate that the proposed method significantly outperforms standard fine-tuning and alternative noise-filtering techniques across multiple benchmark datasets.

0 citationsRead paper
Recent publications

Latest Papers

Tackling the Inherent Difficulty of Noise Filtering in RAG

Jan 05, 2026arXiv.org

This work addresses the vulnerability of retrieval-augmented generation (RAG) systems to performance degradation and hallucination caused by irrelevant or noisy retrieved documents. To overcome the limitations of existing approaches in effectively filtering such noise, the paper proposes a novel fine-tuning strategy that transcends conventional constraints on attention architecture. Specifically, the method introduces a tailored training objective and targeted modifications to the attention mechanism to enhance the model’s ability to discriminate between relevant and irrelevant retrieved content. This approach substantially improves the model’s capacity for information filtering and robustness in noisy retrieval settings. Experimental results demonstrate that the proposed method significantly outperforms standard fine-tuning and alternative noise-filtering techniques across multiple benchmark datasets.

0 citationsRead paper