Elisa Desiato, University of Ferrara

Human–large carnivore coexistence remains a central challenge in conservation. In Central Italy, the Apennine brown bear (Ursus arctos marsicanus) represents a particularly relevant case for investigating long-term dynamics of public perception, as it is now associated with a relatively positive attitude among local populations compared to other European contexts.
This study applies computational text analysis to reconstruct the temporal evolution of attitudes towards the species using historical and contemporary newspaper retrieved from digital repositories and local press archives. The analysis spans from the establishment of the Abruzzo, Lazio and Molise National Park to the present.
Documents were digitized using Optical Character Recognition (OCR) and processed through automated filtering procedures based on keywords and lexical patterns to identify bear-related content.
Sentiment classification was conducted using a Vision-Language model (Qwen2-VL), a multimodal large language model capable of integrating visual and textual information and directly analyzing digitized newspaper pages. A structured prompt guided the model to evaluate: (i) textual relevance to the bear as a real animal, (ii) sentiment polarity (positive, negative, neutral), and (iii) supporting evidence sentences.
Although results are preliminary, the workflow enabled reconstruction of a historical sentiment trend, demonstrating the feasibility of integrating AI-based language analysis with wildlife perception research. This work highlights the potential of computational approaches for examining human–wildlife relationships across extended timescales.