Retrieval-augmented generation (RAG) has attracted significant attention for enhancing large language models (LLMs) in domain-specific and knowledge-intensive tasks by utilizing external documents retrieved by retrievers. However, LLMs often struggle to determine which retrieved documents are relevant and how they relate to one another. We argue that this difficulty arises from a semantic gap between retrievers and LLMs due to differences in their training objectives and architectures. Existing methods either align retrievers and LLMs through costly fine-tuning or feedback signals, but they still do not explicitly strengthen document relationship modeling during generation. This paper proposes ConRAG, a novel enhanced RAG framework to establish an information connection between retrievers and LLMs in RAG, thereby enhancing relationship modeling of LLMs. Specifically, ConRAG employs a lightweight Con-Former model, placed between a retriever and an LLM to capture and transmit semantic information. Then, a semantics injection strategy is employed to integrate the semantic information into the LLM's generation. Accordingly, we employ three tasks for feature modeling and alignment: two relationship modeling tasks to consolidate the local and global perceptions of document relevance and one generative alignment task to facilitate the interpretation of LLM. Notably, ConRAG is suitable for low-resource scenarios where LLMs and retrievers are frozen. Further analysis shows that retriever-derived information helps the LLM better identify relevant evidence and model relationships among documents, leading to more effective generation. The source code is available at https://github.com/yefd/ConRAG.
Fu-Da Ye, Shuang-Yin Li, Yong-Qi Zhang et al.· ACM Transactions on Knowledg...· 0 citations
Drug-drug interaction (DDI) event prediction is critical for ensuring patient safety and optimizing therapeutic outcomes. Existing computational approaches are limited by their inability to jointly model the heterogeneous mechanisms underlying DDIs, which span molecular structure, pharmacodynamic function, and network-mediated relations. To address this limitation, we introduce M2DDI, a unified framework for dynamic multimodal fusion in DDI prediction. M2DDI utilizes a Mixture-of-Experts architecture, with each expert dedicated to a distinct pharmacological modality. A novel prior-enhanced dual-path gating strategy adaptively selects relevant experts for each drug pair by integrating mechanism-matched feature queries and ATC-based biomedical priors, thereby aligning expert selection with underlying pharmacological mechanisms and addressing the challenge of data incompleteness. Empirical evaluation on benchmark datasets demonstrates that M2DDI achieves state-of-the-art performance, particularly in new drug scenarios. Additional robustness experiments show that M2DDI maintains high predictive accuracy even when modality-specific information is partially missing, outperforming existing methods under similar conditions. Analysis of expert selection patterns further confirms alignment with established pharmacological mechanisms. These results establish M2DDI as an effective and mechanism-aware solution for comprehensive DDI prediction. The code is available at: https://github.com/RunqingXuCn/M2DDI.
Runqing Xu, Siyi Liu, Haoyang Li et al.· Proceedings of the 32nd ACM...· 0 citations
Empirical evaluation and robustness experiments show that M2DDI maintains high predictive accuracy even when modality-specific information is partially missing, outperforming existing methods under similar conditions and establish M2DDI as an effective and mechanism-aware solution for comprehensive DDI prediction.
Runqing Xu, Siyi Liu, Haoyang Li et al.· Proceedings of the 32nd ACM...· 0 citations
MolGlass is proposed, a data-centric paradigm for visual molecular understanding in vision-language models (VLMs) that injects chemical priors directly into the visual input through chemical-aware visual augmentations, without modifying model architectures or training molecule-specific encoders.
Runqing Xu, Xiaotang Wang, Chunfeng Gao et al.· Proceedings of the 32nd ACM...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
