[Objective] In the context of climate change, increased production risks, the digital reconfiguration of the allocation of agricultural factors, and the current challenges of food security, green development, and rural revitalisation, traditional agricultural economics research based on small-sample survey, the aggregated statistics, and the linear causal inference has shown decreasing ability to describe nonlinear relationships, spatial-temporal heterogeneity and dynamic interactions in agricultural production and rural society. This research aims to explore the paradigm shift of agricultural economics research catalyzed by big data technologies, explain the implications of big data for the discipline's evidence foundation, analytical thinking and organizational forms, and to draw out the key analytical, data-governance, and ethical problems that come with this shift. It further intends to provide theoretical references for constructing an independent agricultural economic knowledge system with Chinese characteristics amid the ongoing digital transformation of agriculture. [Methods] Drawing on Kuhn's theory of scientific paradigms, the agricultural economics research paradigm is a historically situated ensemble of common commitments, data conventions, and analytical practices and disciplinary norms that guided research. The historical trajectory of agricultural economics in China was traced, and the big-data-driven transformation was examined across seven dimensions: data source, research carrier, research object, research approach, research method, evaluation standard, and research organization. The analysis focused on the way the construction of digital infrastructure and the rise of intelligent analytics transformed agricultural knowledge production. Salient risks attending the deployment of big data were also identified, including biases in data, loss of capacity for causal identification, algorithmic opacity, and new challenges in the field of research ethics and governance. [Results and Discussions] Chinese agricultural economics has experienced four eras of evolution: the early theoretical importation and localization period, the rise of quantitative and econometric analysis period, computationally-assisted inquiry period, and data-intensive scientific inquiry period. The evidentiary base of the field has been transformed, as a massive volume of multi-source data, derived from different technologies, has emerged. Research that used to be limited to small scale, low frequency, structured datasets is now increasingly based on large-scale streams of high-frequency, multi-modal, near real-time data, greatly extending the range and granularity of empirical research. In addition to household surveys and field observations, a hybrid virtual-physical platform, such as digital twin systems, platform-based monitoring and integrated data observatories, was now introduced. The focus expanded from a relatively homogeneous group of farm households to a much more diverse group of small farmers, agribusinesses, actors in the digital platform, rural service providers, and policy actors. Research questions, which were mainly related to static production decisions, were increasingly focusing on bounded rationality, long-term dynamics of behaviour and the ecological-economic-social interactions of rural transformation. Research strategies became more future-oriented, combining forecasting, the modeling of scenarios and cycles of theory revisions. At the method level, machine learning, double machine learning, causal forests, and other tools started to complement standard econometrics, allowing to expand the toolkit to predictive modeling, heterogeneity analysis and more sophisticated causal inference. Evaluative standards were revised also: data quality, predictive fidelity, out-of-sample validity and method-problem fit grew in importance relative to theoretical parsimony and statistical significance alone. Research shifted from isolated, discipline-based research teams to open, collaborative research ecosystems, based on interdisciplinary collaborations and data-sharing platforms. This change was accompanied by four challenges. First, issues of representativeness of data at the data level arose as digital cleavages and a false sense of comprehensiveness often underrepresented remote areas, elderly, and digitally ignored smallholders. Second, at the methodological level, the correlation-causation dichotomy was more prominent since machine learning has been more successful at prediction than in identifying generative mechanisms. Third, algorithmic opacity led to issues of transparency, accountability and fairness in various contexts, including in agricultural subsidy allocation, credit scoring and risk assessment. Fourth, there were increasing risks of privacy infringement and data insecurity, as agricultural big data can also include sensitive personal, operational and strategic data. [Conclusions] Big data technology is driving a transition of agricultural economics from a purely model-centric paradigm towards a data-model co-driven "fourth data-intensive scientific paradigm". It reframes the questions addressed by the discipline, the assumptions it makes, the analytical frameworks it utilizes and the organization of research. Future efforts should focus on four aspects: establishing national unified agricultural big data infrastructure, the integration of data resources, analytical models and algorithmic tools into a shared infrastructure, training and research activities to be undertaken in an interdisciplinary fashion, and constructing governance mechanisms and ethics for agricultural big data.