Deep Learning-Based NLP Framework for Public Sentiment Analysis on Green Consumption: Evidence from Social Media

Green consumption (GC) are crucial for achieving the Sustainable Development Goals (SDGs). However, few studies have explored public attitudes toward GC using social media data, missing potential public concerns captured through big data. To address this gap, this study collects and analyzes public attention toward GC using web crawler technology. Based on the data from Sina Weibo, we applied RoBERTa, an advanced NLP model based on transformer architecture, to conduct fine-grained sentiment analysis of the public’s attention, attitudes and hot topics on GC, demonstrating the potential of deep learning methods in capturing dynamic and contextual emotional shifts across time and regions. Among the sample (N = 188,509), 53.91% expressed a positive attitude, with variation across different times and regions. Temporally, public interest in GC has shown an annual growth rate of 30.23%, gradually shifting from fulfilling basic needs to prioritizing entertainment consumption. Spatially, GC is most prevalent in the southeast coastal regions of China, with Beijing ranking first across five evaluated domains. Individuals and government-affiliated accounts play a key role in public discussions on social networks, accounting for 45.89% and 30.01% of user reviews, respectively. A significant positive correlation exists between economic development and public attention to GC, as indicated by a Pearson correlation coefficient of 0.55. Companies, in particular, exhibit cautious behavior in the early stages of green product adoption, prioritizing profitability before making substantial investments. These findings provide valuable insights into the evolving public perception of GC, contributing to the development of more effective environmental policies in China.