Research On User Travel Route Recommendation Method

Since the concept of"intelligent transportation"was put forward,intelligent travel began to gain the public’s love.Traditional The travel route recommendation algorithm usually generates the shortest or fastest route by the improved weighting algorithm,without considering the user’s travel preference and actual traffic conditions.Users usually do not follow the recommended route in the real world.Therefore,how to provide users with the optimal travel route has become one of the most popular topics for researchers.The existing route recommendation algorithm based on trajectory pays most attention to the trajectory when considering the user’s travel route planning,and relies too much on the user’s historical trajectory data.When encountering the problem of relatively sparse track data,unsatisfactory routes are often recommended.In order to recommend routes that meet users’travel preferences,this paper proposes a route recommendation algorithm SPR~2(Structure-aware,Preference-based Route Recommendation,SPR~2)that senses both structure and preference.The basic idea is to embed travel preference and road network structure characteristics into the representation of each node through multiple self-attention mechanism and Node2vec respectively.Then,a Multilayer Perceptron(MLP)is used to model the mapping of two node representations to the heuristic cost h(·)between them.Finally,h(·)heuristic search path based on A*algorithm.Since SPR~2algorithm needs to calculate all heuristic values of generated child nodes when conducting path search according to A*,resulting in low efficiency of the algorithm,this paper adopts the method of combining reinforcement learning and inverse reinforcement learning to optimize the process of A*search,and proposes a new path recommendation algorithm RLR~2.This method can reduce the search time and improve the efficiency of SPR2algorithm on the basis of ensuring the recommendation performance.This paper uses Geolife and T-Drive data of two groups of users’travel trajectory in Beijing to conduct experiments.Experiments show that the two methods presented in this paper are superior in accuracy,robustness and query efficiency respectively.