May 2025 | New paper entitled "Public sentiment analysis of data center energy consumption using social media data and large language models" was published in *Energy and buildings*.

The full paper can be found here.

Abstract

Data centers play a crucial role in the modern digital industry, but their significant energy consumption, representing about 1.3% of global annual electricity use in 2023, poses a persistent challenge to sustainable development. While extensive research has focused on improving data center energy efficiency through technological innovations, the roles of governmental, social, economic, and legal contributions remain underexplored. Understanding public sentiment can help stakeholders develop holistic strategies that go beyond technology. Traditional survey-based public sentiment analyses are often time-consuming and labor-intensive with limited representativeness and dynamic scope due to small sample size, short durations, and spatial constraints. Research using social media data (SMD) addresses these limitations but faces challenges such as the need for extensive annotated datasets for training and a lack of accuracy or granularity in topic classification. In this study, we aim to develop a holistic approach to analyzing public sentiment and topics related to data center energy consumption using SMD and large language models (LLMs). We collected and preprocessed 104,624 pieces of SMD from Twitter, focusing on data center energy consumption in the USA. Sentiment labels were assigned using three LLMs, with majority voting employed to reduce individual LLM bias and identify the model with the highest F-score (0.953), Gemma 2, for subsequent topic classification. We observed an 87.4% increase in public attention to data center energy consumption in Q4 2022 and identified a predominant shift in sentiment from positive to negative, correlated with average electricity prices in US cities. We proposed a region-specific, three-tier public topic framework to analyze public topic distribution and temporal trends, resulting in four targeted policy recommendations focused on society, economy, and environment (SEE), renewable energy utilization, data center site selection, and economic factors. This study extends insights beyond technical aspects to promote sustainable energy efficiency in data center development and introduces a transferable approach applicable to public sentiment and topic analysis in other regions.