When a tropical cyclone threatens several ports, planners must anticipate which ports will be affected and for how long. Limited historical records at individual ports and uncertainty in landfall location make such estimates difficult. This work introduces a two‑stage recommendation framework: the first stage retrieves candidate ports based on proximity to the expected landfall point, and the second stage ranks those candidates by the estimated number of impact days. Using AIS‑derived impact days for 37 ports and 31 cyclones in the Gulf of Mexico, we evaluate three ways to represent port‑cyclone interactions: pairwise feature interactions, tree‑based feature combinations, and explicit feature crosses with learned embeddings. Tests on Hurricanes Nicole and Ian show that the retrieval stage consistently identifies all observed affected ports despite landfall uncertainty. Among ranking models, Wide & Deep Learning matches the observed port rankings most closely, improving mean ranking agreement by 85.5% for Nicole and 55.5% for Ian relative to a distance‑based baseline, and it is less sensitive to landfall uncertainty. These rankings can help port authorities prioritize post‑storm inspections and allocate recovery resources efficiently. Review