"Research on the Impact of Large Model Technology Diffusion on the Innovation Ecology of the Industrial Chain" - Dataset

This article takes Chinese A-share listed companies as the research object and constructs a panel dataset from 2020 to 2024. The selection during the sample period is based on the following considerations: in 2020, the 100 billion parameter level model represented by GPT-3 was launched, ushering in a new stage of general artificial intelligence, and Chinese enterprises began to systematically explore the industrialization path of large-scale model technology; Using 2024 as the endpoint of the sample not only ensures the timeliness and completeness of the data, but also fully captures the complete evolution process of large-scale model technology from initial exploration to large-scale application. Data collection involves multiple dimensions and sources. The financial information and corporate governance data of listed companies mainly come from the CSMAR database, China Research Data Service Platform (CNRDS), and Wind financial terminal. The relevant data of the application of big model technology is manually collected through multiple channels, including systematically sorting out the relevant disclosure information in the annual reports of listed companies, tracking the technical white papers released by enterprises, and monitoring the participation records of mainstream open source communities. Patent cooperation and citation data are obtained through the patent retrieval system of the China National Intellectual Property Administration. The industrial chain correlation is constructed based on the input-output table released by the National Bureau of Statistics. In the process of sample screening, this article adopted strict data quality control standards: excluding company samples that have been specially processed (ST, * ST); Delete observations with missing or significantly abnormal values in key variables; Exclude companies with only a single annual observation value. All regression models control for fixed year effects to absorb the time trend impact of exogenous shocks such as public health events. After the above processing, 15342 annual observations of enterprises were finally obtained.