LLM-powered snippet method expands supply chain visibility 9x in China

Financial and economic research often relies on structured supply-chain disclosures, but in China, such data is limited to major partners of listed firms. Unlisted firms and long-tail inter-firm links are poorly captured. To address this gap, researchers Fukada and Mizuno introduce a snippet-driven method that leverages web search snippets—query-biased summaries—as a scalable first-pass evidence layer for LLM-based relationship extraction. Compared to exhaustive full-text chunking, which discovers 19.8x more unique relationships, their method uses 251.2x fewer input tokens and significantly reduces redundancy. This trade-off makes large-scale discovery feasible without the prohibitive cost of processing entire web pages.

For evaluation, the team used 130,685 Chinese firms as search seeds, covering both Shanghai/Shenzhen-listed firms and large unlisted firms. The resulting SCKG (supply chain knowledge graph) covered 7.2x more firms and 9.3x more relationships than the CSMAR disclosure-based benchmark. The graph also revealed heavy-tailed degree patterns typical of real-world networks. Crucially, the method retains provenance metadata, making the SCKG an auditable complement to traditional databases. This research opens the door for more comprehensive economic network analysis in regions with limited structured data.