Structured data on inter‑firm relationships is often scarce because many collaborations are privately negotiated, selectively disclosed, and scattered across proprietary databases. This hampers startups and SMEs from uncovering collaboration opportunities. While firm profiles are readily available, business similarity alone cannot infer collaboration potential—similar firms may be competitors, whereas dissimilar firms can offer complementary products, technologies, channels, capabilities, or capital. To address this, we introduce FirmCORe (Inter‑Firm Collaboration Opportunity Reasoning), a human‑annotated benchmark comprising 2,805 labeled firm pairs. The task requires a model, given two firm profiles, to decide whether evidence supports a collaboration opportunity; for positive pairs it must jointly predict opportunity strength, primary collaboration type, and role direction. FirmCORe also provides parallel Chinese and English evaluation sets with identical instances and gold labels, enabling controlled analysis of input‑language sensitivity. Experiments with representative locally deployed and hosted large language models (LLMs) show the strongest model achieves a macro‑F1 of 74.51% for opportunity detection but only 61.57% exact match across all four output fields. Language effects vary across models, and high cross‑language agreement can mask shared errors. These findings indicate that current LLMs are considerably more reliable at detecting broad collaboration opportunities than at pinpointing specific types and role directions.
Review