Large language models (LLMs) excel in various natural language processing tasks and are increasingly applied in specialized fields like medicine. However, their deployment in the medical domain is challenged by limited domain-specific data and the tendency to generate inaccurate information, known as “hallucinations.” While domainspecific fine-tuning has improved open-source LLMs, they still underperform compared to proprietary models like ChatGPT and PaLM. To address this gap, retrieval-augmented generation (RAG) techniques have been explored to enhance LLMs by integrating external knowledge bases. Nevertheless, the success of RAG depends on the quality of retrieved documents, and its application within the medical field remains in the early stages. In this paper, we introduce the “Bailicai” framework as an exploratory approach to integrating RAG with LLMs in the medical field. The framework employs fine-tuning to improve the RAG process, where “falsely relevant” and “completely irrelevant” interference documents are intentionally included in the training data. This enables Bailicai to develop the ability to assess the quality of retrieved documents and selectively incorporate them. The framework is organized into four modules: (1) medical knowledge injection, (2) self-knowledge boundary identification, (3) directed acyclic graph task decomposition, and (4) retrieval-augmented generation. Through the synergy of these modules, Bailicai achieves superior performance on multiple medical benchmarks, outperforming existing large models in the medical domain, RAG-based methods, and proprietary models such as GPT-3.5. Furthermore, Bailicai effectively mitigates the hallucination problem common in LLMs applied to medical tasks and enhances the robustness of RAG when dealing with irrelevant or misleading documents, enabling more accurate information retrieval and integration.

Few-shot Named Entity Recognition (NER) systems are designed to identify new categories of entities with a limited number of labeled examples. A major challenge encountered by these systems is overfitting, particularly pronounced in comparison to tasks with ample samples. This overfitting predominantly stems from spurious correlations, a consequence of biases inherent in the selection of a small sample set. In response to this challenge, we introduce a novel approach in this paper: a causal intervention-based method for few-shot NER. Building upon the foundational structure of prototypical networks, our method strategically intervenes in the context to obstruct the indirect association between the context and the label. For scenarios restricted to 1-shot, where contextual intervention is not feasible, our method utilizes incremental learning to intervene at the prototype level. This not only counters overfitting but also aids in alleviating catastrophic forgetting. Additionally, to preliminarily classify entity types, we employ entity detection methods for coarse categorization. Considering the distinct characteristics of the source and target domains in few-shot tasks, we introduce sample reweighting to aid in model transfer and generalization. Through rigorous testing across multiple benchmark datasets, our approach consistently sets new state-of-the-art benchmarks, underscoring its efficacy in few-shot NER applications.