MM-PRS: A multi-modal parking recommendation system based on extracting parking features and deep learning

Abstract

Abstract

In contemporary urban environments, the task of finding suitable parking is a significant challenge. This study proposes the Multi-Modal Parking Recommendation System (MM-PRS) aiming to enhance user satisfaction for personalized parking suggestions in large cities. The proposed MM-PRS functions in three stages. Initially, the Parking Feature Extraction (PFE) stage extracts pertinent features by analyzing user parking records, along with environmental and temporal factors. It ensures a deep understanding of user needs and parking patterns. Second, the User Clustering (UC) stage utilizes unsupervised machine learning methods to group users based on the extracted features. The users in the same group can apply the same model to recommend their parking lot. Finally, the Parking Lot Recommendation (PLR) stage employs multi-modal architecture to provide tailored parking lot suggestions. This stage comprises three integrated models. The first is a supervised machine learning model, which performs coarse-grain classification of parking lots based on user satisfaction metrics and extracted features. Following this, a compatibility filtering model assesses parking lots against user criteria, such as distance, time, and availability, to generate a list of candidates for parking lots. Lastly, a deep learning model analyzes these parking lots alongside user features to make the final parking recommendations. Experimental results demonstrate that MM-PRS significantly outperforms existing systems in Precision, Recall, and F1-Score, showcasing its effectiveness in navigating the complexities of urban parking scenarios. By integrating satisfaction estimation into the recommendation process, MM-PRS ensures that the suggested parking lots align with individual user preferences.