A Heuristic Framework for Personalized Route Recommendation Based on Convolutional Neural Networks

Abstract

Personalization of travel routes significantly impacts people’s quality of life and production efficiency. The personalized route recommendation (PRR) problem has personalized requirements and the goal of providing users with personalized route suggestions. Most of the existing work focuses on improving either the personalization or the availability of recommended routes, rather than both. In response to the above problems, a Personalized-Neural-Network-Heuristic framework (PNNH) is proposed, which can improve the personalization degree of the recommended routes and ensure their availability simultaneously. The PNNH framework consists of two stages: preference modeling and route recommendation. In the preference modeling stage, a prediction component with Graph Convolutional Network (GCN) as the core is constructed to learn the potential preference characteristics in the user’s historical travel information, and then a heuristic algorithm is constructed by using the evaluation value output by the prediction component reflecting the transition probability, thus introducing the user preference characteristics into its cost evaluation function. In the route recommendation stage, an improved heuristic algorithm is used for route planning, and the route planning results are recommended to users. A strategy of narrowing the search scope for the heuristic algorithm is proposed, which can ensure that the route reaches its destination. Based on the PNNH framework, a set of algorithms can be constructed. The NeuroMLR-Dijkstra-A* algorithm (NDA*) is constructed and used in the experiment to evaluate the performance of the PNNH framework. Experimental results demonstrate the superiority of the PNNH framework.