Empowering A* Search Algorithms with Neural Networks for Personalized Route Recommendation

Abstract

Personalized Route Recommendation (PRR) aims to generate user-specific route suggestions in response to users' route queries. Early studies cast the PRR task as a pathfinding problem on graphs, and adopt adapted search algorithms by integrating heuristic strategies. Although these methods are effective to some extent, they require setting the cost functions with heuristics. In addition, it is difficult to utilize useful context information in the search procedure. To address these issues, we propose using neural networks to automatically learn the cost functions of a classic heuristic algorithm, namely A * algorithm, for the PRR task. Our model consists of two components. First, we employ attention-based Recurrent Neural Networks (RNN) to model the cost from the source to the candidate location by incorporating useful context information. Instead of learning a single cost value, the RNN component is able to learn a time-varying vectorized representation for the moving state of a user. Second, we propose to use a value network for estimating the cost from a candidate location to the destination. For capturing structural characteristics, the value network is built on top of improved graph attention networks by incorporating the moving state of a user and other context information. The two components are integrated in a principled way for deriving a more accurate cost of a candidate location. Extensive experiment results on three real-world datasets have shown the effectiveness and robustness of the proposed model.

Personalized Route Recommendation With Neural Network Enhanced Search Algorithm

“…Our current study significantly extends previous work [25] in multiple aspects. First, we enhance the H(•) function (the key of the A * search algorithm) for estimating the future cost with a totally new position-aware GNN.…”

“…• NASR [25]: it is a base version of our model, which shares the same component for the G(•) function but adopts a simpler implementation of the H(•) function without considering position-aware node embeddings.…”

“…With the availability of user-generated trajectory information, route recommendation has received much attention from the research community [3], [5], [6], [23], [24], which aims to generate reachable paths between the source and destination locations. The task can be defined as either personalized [5], [6], [25] or nonpersonalized [8], [10], [23], [26], and constructed based on different types of trajectory data, e.g., GPS data [26], [27] or POI check-in data [28], [29]. In the literature, various algorithms have been developed for route recommendation.…”

“…More recently, deep learning has significantly pushed forward the research of this line. The main idea is to leverage the powerful modeling capacity of neural networks for improving the tasks that require complicated solving strategies, including the Go game [17] and route recommendation [25]. Our work is highly inspired by these pioneering works, but have a quite different focus on the studied task, i.e., personalized route recommendation.…”

“…In 1. We use the suffix of the plus symbol in naming our model for discriminating this work from our KDD paper [25].…”