大语言模型领域意图的精准性增强方法.

Large language models (such as GPT) exhibit instability and inauthenticity in professional domain Q&A applications. To address this issue, this paper proposed a method to enhance intent recognition by domain knowledge (EIRDK) for large language models. The method involved three specific strategies: a) scoring and filtering the GPT output using a domain knowledge base, b) training the domain knowledge word vector mode to optimize prompt, e) utilizing feedback from GPT to improve the coherence between the domain word vector model and the GPT model. Experimental analysis demonstrates that, compared to the standard GPT model, the new method achieves a 25% improvement in intent understanding accuracy on the private dataset and a 12% increase on the CMID dataset. The results validate the effectiveness of the EIRDK method. [ABSTRACT FROM AUTHOR]