PROJECT TITLE :

Incorporating Multi-Source Urban Data for Personalized and Context-Aware Multi-Modal Transportation Recommendation

ABSTRACT:

The recommendation of appropriate modes of transportation is an essential component of map services provided by navigation applications. Previous solutions for transportation recommendation have not been successful in providing a satisfactory user experience. This is due to the fact that their recommendations only consider routes in a single mode of transportation (uni-modal, for example, taxi, bus, or cycle) and largely ignore the situational context. In this work, we propose the system known as $mathsf 'Hydra$, which is a multi-task deep learning based recommendation system that offers multi-modal transportation planning and is adaptable to various situational contexts (for example, the distribution of nearby points of interest (POI) and the weather). We design a novel two-level framework that integrates uni-modal and multi-modal (e.g., taxi-bus, bus-cycle) routes as well as a variety of urban data in order to make intelligent multi-modal transportation recommendations. This is accomplished by capitalizing on the availability of existing routing engines and large amounts of urban data. We learn the latent representations of users, origin-destination (OD) pairs, and transportation modes based on user implicit feedbacks. This captures the collaborative transportation mode preferences of users and OD pairs and complements the urban context features that are constructed from multiple sources of urban data. In addition, we propose two models to recommend the best route among the various uni-modal and multi-modal transportation routes: (1) a light-weight gradient boosting decision tree (GBDT) based recommendation model, and (2) a multi-task wide and deep learning (MTWDL) based recommendation model. Both of these models are based on the idea that the optimal route is the one that combines several different modes of transportation. In addition, we optimize the framework to support large-scale real-time route querying and recommendation.