Algorithm Optimization and Performance Improvement of Debt Enterprise Information Retrieval System in the Big Data Environment

Abstract

Against the backdrop of the big data era, the debt enterprise information retrieval system, as the core tool for financial risk management, is confronted with the challenge of processing massive heterogeneous data. The multi-source heterogeneity, high-frequency dynamics and concealed correlations of debt information lead to high data integration costs, difficult timeliness guarantee and insufficient penetration of deep risks, causing deviations in risk assessment and errors in the prediction of innovation potential. This paper reviews the existing technical solutions, deeply analyzes the characteristics of debt information and industry problems, including the shortcomings of insufficient capture of debt dynamics and lack of adaptability, discusses the combined optimization of distributed architectures such as Hadoop-ElasticSearch for massive data processing, and the hybrid BM25- neural embedding framework to improve the accuracy and efficiency of retrieval. As well as the application effectiveness and limitations of the RAG model in a multilingual environment, and evaluate key constraints such as insufficient intention understanding, weak cross-modal fusion, and lack of time sensitivity. The significance of this article lies in providing a comprehensive reference basis for the improvement of system performance, supporting real-time and accurate risk assessment and policy response, and indicating future research directions, including deepening the intention understanding model, developing cross-modal fusion mechanisms, and optimizing temporal dimension filtering.