An LLM-assisted workflow for automated literature analysis

Abstract

The growing volume and diversity of scientific literature pose significant challenges for systematic analysis. While systematic reviews and mapping approaches provide structured methodologies, they remain largely manual and difficult to scale. At the same time, advances in large language models (LLMs) and scholarly data infrastructures enable new opportunities for automating literature analysis. However, existing approaches often focus on isolated tasks rather than integrated workflows. This paper proposes an LLM-assisted workflow for automated literature analysis that integrates corpus construction, hybrid relevance screening, schema- guided information extraction, and human validation within a reproducible workflow structure. The approach combines automated retrieval via scholarly application programming interfaces (APIs) with a transparent scoring mechanism that integrates keyword and model- based relevance signals, as well as evidence-based extraction with traceable outputs. The workflow is evaluated using literature on automated simulation model generation as an example domain and benchmarked against a structured comparison table from an existing systematic review. The results demonstrate that the workflow can produce consistent and traceable structured outputs while reducing manual effort. At the same time, only a subset of benchmark studies could be processed automatically, reflecting the influence of data accessibility and document structure on reproducibility. Overall, the proposed workflow provides a transparent and reproducible approach to integrating LLMs into literature analysis, supporting scalable knowledge extraction while maintaining human oversight.