Quantifying the higher-order influence of scientific publications