Port Recommendation System for Alternative Container Port Destinations Using a Novel Neural Language-Based Algorithm

Abstract

Shipping containers are tokens of multimodal international transportation and rapid logistics. Container deliveries are scheduled to satisfy rapidly changing requirements. Unpredictable increases in costs and unforeseeable events such as pandemics compel ship owners and managers to adopt risk minimization measures. This study addresses one issue: How to determine an alternative port of call from massive data to offer a realistic destination change recommendation for a container vessel. Recommendation algorithms have become ubiquitous and are used effectively in otherfields, but there is no such model for the port of call selection or recommendation. Large scale automatic identification system (AIS) data are readily available. We developed a computational framework based on a novel natural language programming algorithm that was tailored to support port recommendation rather than use a conventional adjacency matrix method. We mined large scale AIS data to construct sequential berth records for container vessels and mapped each port onto a vector in an embedded space. The natural language neural programming algorithm can suggest ports similar to the scheduled ports of call that were unable to offer service. The recommendations were validated with geo-analysis of sailing distance and could offer viable alternative ports to shipping managers.