This study investigates a method that integrates retrieval-augmented mechanisms into large language model agents for scientific literature review generation. The approach addresses the limitations of traditional review models that rely on parametric knowledge with insufficient timeliness and limited coverage. Incorporating external document retrieval and dynamic information fusion into the generation process enhances the accuracy and completeness of the output. The overall framework consists of query encoding, semantic retrieval, document filtering, knowledge fusion, language modeling, task planning, memory storage, and reinforcement optimization, forming a closed loop of retrieval, understanding, and generation. Relevant document fragments are first retrieved through semantic vector search to ensure comprehensive and reliable information sources. These external representations are then integrated with the internal embeddings of the language model through weighted fusion, which preserves fluency while maintaining factual grounding. The task planning module constrains logical flow and text structure, and reinforcement learning optimization further improves relevance and consistency. Comparative experiments on large-scale scientific literature datasets demonstrate that the method outperforms existing approaches on ROUGE, BLEU, METEOR, and diversity metrics, validating its effectiveness and practicality. The findings show that combining retrieval augmentation with agent architectures can significantly improve coverage, accuracy, and language quality in review generation, providing a feasible solution for knowledge organization in complex literature environments.

This paper focuses on the efficient fine-tuning of large language models within the federated learning framework. To address the performance bottlenecks caused by multi-source heterogeneity and structural inconsistency, a structure-aware federated fine-tuning method is proposed. The method incorporates a graph representation module (GRM) to model internal structural relationships within text and employs a segmentation mechanism (SM) to reconstruct and align semantic structures across inputs, thereby enhancing structural robustness and generalization under non-IID (non-Independent and Identically Distributed) settings. During training, the method ensures data locality and integrates structural pruning with gradient encryption (SPGE) strategies to balance privacy preservation and communication efficiency. Compared with representative federated fine-tuning baselines such as FedNLP and FedPrompt, the proposed method achieves consistent accuracy and F1-score improvements across multiple tasks. To evaluate the effectiveness of the proposed method, extensive comparative experiments are conducted across tasks of text classification, named entity recognition, and question answering, using multiple datasets with diverse structures and heterogeneity levels. Experimental results show that the proposed approach significantly outperforms existing federated fine-tuning strategies on most tasks, achieving higher performance while preserving privacy, and demonstrating strong practical applicability and generalization potential.