Leveraging Large Language Models and Social Media for Power Outage Detection

Social media is increasingly used in various emergency response scenarios, such as natural disasters, public health crises, and security incidents. This study explored the potential to overcome the limitations of traditional Outage Management Systems (OMSs) data collection by leveraging the widespread use of social media and advancements in Machine Learning (ML). The objective of this study is to propose and validate a multi-step processing and interaction pipeline that could filter, analyze, and extract relevant information from social media posts, ultimately integrating the obtained information into OMSs for more efficient outage management. Our validation process revealed that large language model Mixtral- $8 \times 7 \mathrm{~B}$-Instruct-v 0.1 can accurately flag posts related to ongoing power outages with a 91% success rate. By integrating ML-driven insights from social media, utility companies could improve their response capabilities, foster community engagement, and build a more resilient power infrastructure.