Machine Learning Assessment of Community Attitudes Toward Carnivore Conservation

Abstract

The growing role of public acceptance, especially among people who live in the vicinity of predators' habitats, is becoming very significant in carnivore conservation. While conventional surveying and interviewing techniques offer information about community perceptions, are expensive, geographically constrained and cannot detect fluctuations of attitudes after livestock losses, wildlife encounters or policy changes. This research builds a machine learning platform to analyze attitudes in communities towards carnivore conservation based on digital discourse that is collected from social media platforms, online forums, and news comment sections. The textual data is processed using tokenization, stop-word removal, lemmatization and represented using TF-IDF features and contextual embeddings. Supervised transformer-based classification is used to detect sentiment (positive, neutral, negative) while unsupervised topic modelling is used to identify themes that are not labelled. Four major attitude clusters are identified: ethical preservation, economic anxiety, governance distrust and coexistence optimism. Urban discourse indicates general level of abstraction and support for carnivores, while rural discourse indicates concern on the losses of livestock, personal safety, compensation, and livelihood security. The institutional trust is identified as a determinant of whether the discontent is aimed at carnivores or at the wildlife management authorities. Results showed that machine learning could identify spatially and emotionally different patterns which might not be revealed in periodic surveys. The proposed approach has the potential of providing real-time sentiment monitoring, local communication and participatory conservation planning. But not all communities are represented in digital discourse and there might be sarcasm, cultural ambiguity, and demographic bias. Thus, the results of computation should be used in conjunction with field surveys and direct involvement. Future work should confirm predictions with community-level survey data and look for attitude shifts over time for specific conservation interventions.