It naturally ranks items by predicted preference ( \hatr_ui = u_u v_i^\top ), and the confidence weighting prevents over‑counting negative signals.