Energy-centric demand evolution and preference forecasting for new energy vehicles: An EEMD–Attention–BiGRU framework based on user-generated content

Accurately capturing how consumer energy-related priorities evolve is pivotal for guiding the low-carbon transition of road transport. Drawing on more than 47,000 online reviews from China's AutoHome forum, we build a user-generated content-driven, two-stage framework that first fuses BERTopic modeling with aspect-level bidirectional encoder representations from transformers sentiment analysis to extract energy-salient attributes—battery durability, charging convenience, efficiency—and classify them via an extended Kano model that incorporates Mann–Kendall trend testing to reveal whether once-“indifferent” attributes are becoming value-added or obsolete. We then forecast the future trajectories of user satisfaction and attention for each attribute by decomposing the multi-frequency demand signals with ensemble empirical mode decomposition and learning bidirectional temporal dependencies with an attention-enhanced bidirectional gated recurrent unit network, achieving mean-absolute-error improvements of 12–21% over transformer- and BiLSTM-based baselines while retaining near-real-time runtimes (∼10 s per forecast). Empirical results highlight a rapid uplift in sentiment toward high-density battery packs and ultrafast charging—signaling emerging energy “value levers”—while declining satisfaction with cabin space warns of design bottlenecks. The proposed EEMD–Attention–BiGRU pipeline thus converts noisy social feedback into fine-grained, forward-looking intelligence that can steer automakers, grid planners, and policy makers toward more resilient battery strategies, targeted charging-infrastructure roll-outs and agile product road-mapping in the sustainable-energy era.