Data set of pro-policies in four places

The Chroma RAG model used in this study demonstrates significant advantages in Hit@1 The target has reached 60% Hit@3 The indicator reached 76%, and the MRR indicator reached 71.13%. In the comparison of retrieval methods, the performance of our research model is superior to traditional models such as Tf idf, Word2Vec, USE, BERT, SBERT, DPR, SimCSE, etc., highlighting the superiority of our method. By Hit@1 The Hit@3 Compared with evaluation indicators such as MRR, the framework and retrieval method proposed in this study have significant advantages.

[Limitations] The research is mainly based on cross-sectional data and cannot fully reflect the dynamic changes in the policy implementation process, which limits the in-depth analysis of policy effects. [Conclusion] The construction of knowledge base and policy comparison based on big language model can effectively improve the intelligent analysis and comparison effect of policy texts, especially in the construction of policy knowledge base and policy comparison support, providing significant decision support for policy makers.