Communications Materials Механика материалов

An efficient AI-assisted approach for retrieving knowledge of polymer biodegradability from a corpus of publications

Sonakshi Gupta, Akhlak Mahmood, Wei Xiong, Rampi Ramprasad
Springer Nature · 2026-07-15 · DOI 10.1038/s43246-026-01267-x

Аннотация

Abstract Polymer literature contains a large and growing body of experimental knowledge, yet much of it remains buried in unstructured text and inconsistent terminology, limiting systematic retrieval and reasoning. Existing tools typically extract narrow, study-specific facts in isolation, failing to preserve the cross-study context required to answer broader scientific questions. Retrieval-augmented generation (RAG) offers a promising approach by combining large language models (LLMs) with external retrieval, but its effectiveness depends strongly on how domain knowledge is represented. In this work, we develop two retrieval pipelines: a dense semantic vector-based approach (VectorRAG) and a graph-based approach (GraphRAG). Using over 1,000 polyhydroxyalkanoate (PHA) papers, we construct context-preserving paragraph embeddings and a canonicalized structured knowledge graph that supports entity disambiguation and multi-hop reasoning. We evaluate both pipelines using standard retrieval metrics, comparisons with state-of-the-art systems such as GPT and Gemini, and qualitative validation by a domain chemist. GraphRAG achieves higher precision and interpretability, while VectorRAG provides broader recall, highlighting complementary trade-offs. By grounding responses in explicit evidence, the framework enables trustworthy literature analysis at scale. More broadly, this work establishes a practical foundation for building materials science assistants using curated corpora and retrieval design, reducing reliance on proprietary models.

Читать у издателя →