DC-RAG: a dual-channel retrieval-augmented generation framework for audit analysis.

Area of Science:

• Information Retrieval
• Artificial Intelligence
• Knowledge Management

Background:

• Intelligent question-answering systems are increasingly vital for information retrieval.
• Existing models lack domain-specific adaptations for analyzing audit result announcements.
• This gap hinders the effective utilization of structured and semantic information within audit texts.

Purpose of the Study:

• To develop a novel framework for intelligent question-answering specifically for audit result announcements.
• To address the scarcity of domain-specific models in audit text analysis.
• To enhance the exploitation of structured and semantic information in audit data.

Main Methods:

• Proposed a dual-channel retrieval-augmented generation framework (Dual-Channel Retrieval-Augmented Generation).
• Implemented a dual-path retrieval mechanism accessing both document and relational databases.
• Integrated multi-source evidence using a unified evaluation and re-ranking strategy.

Main Results:

• The proposed model demonstrated superior performance compared to baseline methods.
• Achieved higher effectiveness in both information retrieval and answer quality.
• Accurately interpreted user queries and generated high-quality responses.

Conclusions:

• The Dual-Channel Retrieval-Augmented Generation framework offers a feasible solution for intelligent applications of audit result announcements.
• The model enhances audit semantic understanding and reasoning capabilities.
• Successfully enables collaborative knowledge retrieval and answer generation from multiple data sources.