To address the challenges of insufficient domain knowledge coverage and semantic misunderstanding of technical terms when applying retrieral-augmented generation (RAG) to terminology-intensive and logically strict fields such as signal processing, a QA optimization method for signal processing knowledge based on embedding model fine-tuning is proposed. First, a high-quality dataset in the signal domain is constructed, and a contrastive learning strategy is adopted to fine-tune the embedding model in the retriever for better domain adaptation. Subsequently, a RAG system is built using the fine-tuned embedding model. Hybrid retrieval and re-ranking mechanisms are introduced during the retrieval phase, while prompt optimization is applied in the generation phase to enhance the comprehension and quality of outputs related to signal domain knowledge. Experimental results demonstrate that the fine-tuned embedding model significantly improves multiple performance metrics in signal-domain retrieval tasks. The optimized RAG system also outperforms baseline methods in terms of faithfulness, answer relevancy, and contexts precision.
LEWISP, PEREZE, PIKTUSA, et al. Retrieval-augmented generation for knowledge-intensive NLP tasks[J]. Advances in Neural Information Processing Systems, 2020,33:9459-9474.
[4]
GAOX F, XIONGY, GAOX Y, et al. Retrieval-augmented generation for large language models: a survey[DB/OL]. (2023-12-18)[2025-06-10].
[5]
NOGUEIRAR, CHOK. Passage re-ranking with BERT.[DB/OL].(2019-01-13)[2025-06-10].
[6]
ZHENGH X S, MISHRAS, CHENX Y, et al. Take a step back: evoking reasoning via abstraction in large language models[DB/OL]. (2023-10-09)[2025-06-10].
SAWARKARK, MANGALA, SOLANKIS R. Blended RAG: improving RAG (retriever-augmented generation) accuracy with semantic search and hybrid query-based retrievers[C]//Proceedings of the 2024 IEEE 7th International Conference on Multimedia Information Processing and Retrieval. Piscataway, USA: IEEE, 2024:155-161.
[10]
SHAOZ H, GONGY Y, SHENY L, et al. Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy[C]//Proceedings of the Findings of the Association for Computational Linguistics: EMNLP 2023. Stroudsburg, USA: ACL, 2023:9248-9274.
[11]
ASAIA, WUZ Q, WANGY Z, et al. Self-RAG: learning to retrieve, generate, and critique through self-reflection[DB/OL].(2023-10-18)[2025-06-10].
[12]
DUJ J, DINGJ B, LIJ Y, et al. SC-Telco RAG: enhancing RAG through structured contexts in telecommunications[C]//Proceedings of the 2024 IEEE Globecom Workshops. Piscataway, USA: IEEE, 2024:1-6.
BEIY J, FANGZ B, MAOS Y, et al. Manufacturing domain QA with integrated term enhanced RAG[C]//Proceedings of the 2024 International Joint Conference on Neural Networks. Piscataway, USA: IEEE, 2024:1-8.
[15]
XIAOS T, LIUZ, ZJHANGP, et al. C-Pack: packed resources for general Chinese embeddings[DB/OL].(2023-09-14)[2025-06-10].
[16]
CHENJ L, XIAOS T, ZHANGP T, et al. BGE M3-embedding: multi-lingual, multi-functionality, multi-granularity text embeddings through self-knowledge distillation[C]//Proceedings of the Findings of the Association for Computational Linguistics ACL 2024. Stroudsburg, USA: ACL, 2024:2318-2335.
[17]
Github-Infiniflow. RAGFlow is an open-source RAG (Retrieval-Augmented Generation) engine based on deep document understanding[EB/OL]. (2025-02-28)[2025-06-10].
[18]
BAIJ Z, BAIS, CHUY F, et al. Qwen technical report[DB/OL].(2023-09-29)[2025-06-10].
[19]
Deepseek-AI, GUOD Y, YANGD J, et al. DeepSeek-R1: incentivizing reasoning capability in LLMs via reinforcement learning[DB/OL].(2025-01-22)[2025-06-10].
[20]
ES S, JAMESJ, ESPINOSAA L, et al. RAGAs: automated evaluation of retrieval augmented generation[C]//Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics: System Demonstrations. Stroudsburg, USA: ACL, 2024: 150-158.