The Impact of Artificial Intelligence on Customer Personalization Strategies in E-Commerce

Abstract

Against the background of explosive growth in scientific research data and accelerated interdisciplinary integration, large language models (LLMs) have become vital assistants for scientific discovery. However, issues such as knowledge obsolescence, factual hallucination, and citation fabrication seriously restrict their application in high-rigor scientific scenarios. Retrieval-Augmented Generation (RAG) significantly improves the factual foundation and timeliness of generated content by dynamically accessing authoritative external literature, experimental data, and knowledge bases. Trustworthy verification ensures the reliability of scientific conclusions from multiple dimensions: factual consistency, logical rigor, source traceability, and experimental reproducibility. This paper systematically sorts out the technical framework, core components, and optimization paths of LLM-RAG for scientific discovery, and constructs a full-link trustworthy generation system covering retrieval–generation–verification–feedback. It analyzes unique challenges in scientific scenarios, including professional semantic understanding, multimodal data adaptation, ethical compliance, and open science constraints. A hierarchical trustworthy verification mechanism is proposed based on fact-checking, citation alignment, logical verification, and experimental constraints, whose effectiveness is validated with typical cases in drug discovery, materials science, and bioinformatics. Finally, future directions are prospected, including multimodal scientific RAG, adaptive trustworthy verification, human–machine collaborative research paradigms, and ethical supervision systems. This paper provides theoretical references and technical solutions for LLMs to empower scientific discovery safely, reliably, and efficiently.