MINING SENTIMENT IN AGRIBUSINESS NEWS: AN ANALYSIS OF THE CORRELATION WITH PRICES IN THE POULTRY CLUSTER OF BASTOS-SP

Abstract

This study investigates the correlation between the sentiment of specialized media and price formation in the poultry cluster of Bastos-SP, a hub responsible for approximately 11% of Brazilian egg production. The central objective of the work is to validate whether the information flow acts as a leading indicator or whether the local market operates decoupled from digital expectations due to physical fundamentals. The applied methodology, of an exploratory and quantitative nature, employs Natural Language Processing (NLP) techniques through the BERTimbau Deep Learning model to analyze a corpus of headlines collected between 2023 and 2026. The data were subjected to semantic filtering and weekly resampling using the Pandas library's “.resample('W')” method in Python. The results demonstrate that source curation and the removal of technical noise tripled the observed Pearson correlation (from r = 0.04 to r = 0.12), indicating that egg prices are marginally sensitive to macroeconomic factors rather than to zootechnical information. The weak, but positive, correlation (r = 0.12) observed after the inclusion of macroeconomic sources suggests that local producers are marginally more sensitive to structural trends (grain costs, external environment) than to zootechnical technical factors, which proved statistically irrelevant for short-term price forecasting. Therefore, it is concluded that the weak magnitude of the correlation confirms the dominance of the spot market and the biological nature of the asset, positioning media sentiment predominantly as a coincident indicator of market conditions.