Coupled-NeuralHP: Directional Temporal Coupling Between AI Innovation Exposure and Public Response

A new research paper by Amir Rafe and Subasish Das, titled “Coupled-NeuralHP: Directional Temporal Coupling Between AI Innovation Exposure and Public Response,” presents a powerful hybrid event-plus-state model that deciphers the relationship between AI patent activity and public interest. The model links eight-domain USPTO AI patent publication streams to a train-only Google Trends response index, then validates its forecasts against rigorous baselines. The best variant—a one-way innovation-to-response model—delivers held-out innovation count forecasts with a pseudo-log-likelihood of -30.4 (vs. -34.7 for baselines) and an RMSE of 471 (vs. 532), while matching the multi-lag factor-family baseline on response RMSE (0.295).

Ablation studies reveal that the real-data response signal is carried primarily by the structured forecast head; the reverse response-to-innovation block fails to improve held-out count prediction. Across 60 semi-synthetic replications, the Coupled-NeuralHP family recovers innovation-to-response links with an F1 score of 0.734, vastly outperforming vector autoregression with exogenous inputs (VARX) at 0.386. A placebo-controlled split-date analysis (2022) finds no robust milestone-specific regime break, suggesting that the relationship is consistent over time. The paper is available on arXiv (2605.04194) and has practical implications for forecasting AI hype cycles and R&D strategy.