Snippet-Driven Supply Chain Discovery with LLMs: Scaling Visibility in China

A recent study introduces a snippet-based approach for developing a supply chain knowledge graph (SCKG) in China, utilizing web search snippets as a scalable initial evidence layer for relationship extraction via LLMs. This strategy tackles the shortcomings of structured supply chain disclosures, which often only include major partners of publicly listed companies, neglecting unlisted firms and less prominent inter-firm connections. While public web evidence from corporate, governmental, and trade media can help fill this gap, large-scale full-text web mining is expensive due to inaccessible pages and the high costs associated with processing large language models. The proposed method leverages query-biased summaries from search results to efficiently extract inter-firm relationships. The efficiency and coverage of the pipeline are assessed. The study is accessible on arXiv with the identifier 2605.27845.