Published Paper Details:

The choice of a route by driver to a destination may depend on its distance and travel time, also factor into the driver's decision. Driver's route choice can be context dependent, e.g., varies with respect to time, distance and also it varies from driver to driver. Conventional routing services supports less in context dependent route selection, also they deliver same routes to all drivers. Here the study is to identify context-aware routes for each drivers based on historical trajectories. Ranking paths becomes an increasingly most important functionality in many transportation services, where multiple paths connecting a source-destination pair are offered to drivers. First propose a real training data using enriching method to obtain a dense and expanded set of training paths using historical trajectories. Second, a multi-task learning framework that consider features that take into account both candidate paths and contexts are proposed. Third, propose road network embedding which uses road network topology and context-aware properties to embed the paths into feature vectors. Next, propose a path rank method to generate a compact and diverse set of training paths to enable efficient and effective learning. Finally, propose Adaptive Routing Algorithm to shows that the accuracy results outperforms baseline methods.