Collaborative online ranking algorithms for multitask learning

There are many applications in which it is desirable to rank or order instances that belong to several different but related problems or tasks. Although unique, the individual ranking problem often shares characteristics with other problems in the group. Conventional ranking methods treat each task independently without considering the latent commonalities. In this paper, we study the problem of learning to rank instances that belong to multiple related tasks from the multitask learning perspective. We consider a case in which the information that is learned for a task can be used to enhance the learning of other tasks and propose a collaborative multitask ranking method that learns several ranking models for each of the related tasks together. The proposed algorithms operate in rounds by learning models from a sequence of data instances one at a time. In each round, our algorithms receive an instance that belongs to a task and make a prediction using the task’s ranking model. The model is then updated by leveraging both the task-specific data and the information provided by other models in a collaborative way. The experimental results demonstrate that our algorithms can improve the overall performance of ranking multiple correlated tasks collaboratively. Furthermore, our algorithms can scale well to large amounts of data and are particularly suitable for real-world applications in which data arrive continuously.