This study aims to develop an approach for cross-analyzing public opinions on security issues in Burkina Faso, as expressed on social media platforms. We implemented a methodology combining web scraping, API calls, and large language models (LLMs) to collect and analyze data from Fasonet [1], Burkina24 [2] and YouTube [3]. The experiment involved processing 5,189 opinions using GPT-4 and Gemini 1.0 Pro models for sentiment analysis and theme identification. The results showed that GPT-4 achieved superior performance with 99.16% accuracy in opinion classification, outperforming previous methods. The study are revealed significant variations in public perceptions of security throughout 2023 and highlighted YouTube as a dominant platform for public discourse. This multidimensional approach offers a comprehensive understanding of social media opinions, though limitations in data collection and model dependence were identified, suggesting avenues for future research.

Social media analysis , Opinion mining , Sentiment analysis , Large Language Models (LLMs) , Cross-platform analysis