Research on Automatic Generation of Financial Research Reports Based on Large Model Integration and Small Discrimination Models

Abstract

Although large language models have made significant progress, automatic generation of financial research reports is still limited by quality fluctuations of single-model architectures and insufficient domain expertise. Existing research mainly focuses on isolated model applications, while multi-model collaborative mechanisms and modular quality assessment systems for the rigor of financial text structure remain lacking. To address these issues, this paper proposes a dual-layer architecture of “multi-generation models + lightweight discriminative models.” Specifically, heterogeneous large models like DeepSeek, Kimi, and Gemini are used to generate multiple versions of research reports, and prompt design is optimized through retrieval-augmented generation (RAG); Establishing a five-dimensional quality assessment framework, fine-tuning GPT-5-mini based on high-scoring Guosheng Securities research reports and training data from 50 industry research reports, achieving professional alignment across fields; Furthermore, a modular optimization strategy and a cross-model consistency validation algorithm are designed to integrate the strengths of each generator. Main innovations: Establishing a three-layer paradigm of “generation—evaluation—integration,” extending large model integration theory to professional financial text production; A fine-tuning method for domain alignment discrimination models for multidimensional quality assessment is proposed; A modular preferential fusion mechanism is designed to tap into the complementary advantages of heterogeneous models while ensuring structural and logical consistency. This study deepens the theoretical understanding of multi-model collaboration in domain-specific text generation, providing a scalable framework for automated production of financial research reports. The proposed “manual control + intelligent generation” human-machine collaboration protocol clarifies the intervention trigger mechanism under extreme market conditions, balancing automation efficiency with professional prudence.