Bridging supply–demand gap in open data ecosystem: A dual-perspective approach for optimizing provider collaboration and predicting user demand

At present, data has become a new type of production factor. How to fully release the value of data becomes one of the major challenges. Despite the rapid expansion of public data platforms, a mismatch remains between the rigid supply of data and the dynamic demands of users. Existing platforms often operate as passive repositories, lacking effective methods to predict user demands or coordinate cross-departmental collaboration. To address this challenge, this study proposes a data-driven framework that integrates complex network with Large Language Models (LLMs) to facilitate the implementation of a Data Product-Service System (DPSS). On the supply side, to overcome the problem of isolated data provision where departments operate independently without recognizing potential synergies, we model the data provision as a bipartite graph. By employing improved projection methods and the CRITIC-TOPSIS, we quantify the collaboration potential among providers and identify influential leading departments in the ecosystem. This enables cross-departmental data fusion and supports the structural optimization of data supply. On the demand side, we develop an interpretable framework to infer user demands. By employing a knowledge distillation strategy, we demonstrate that lightweight LLMs can effectively accomplish tasks such as semantic extraction, while maintaining strong generalization capabilities. Using fine-tuned LLMs and the developed relative position encoding method, keywords can be extracted from a data catalog and a concept network can be constructed. Then, we propose a novel Value-Biased Probabilistic Walk (VBPW) algorithm to explore potential high-value data products within the network. Finally, we validate our framework by using the Shanghai public data catalog as a case study. This research provides a quantifiable methodological foundation for transforming data supply platforms from passive storage and keyword matching into active demand-centric service provision systems.