Iterative framework based on multi-task learning for service recommendation

Abstract

In recent years, service-oriented computing technology has developed rapidly, which, however, has increased the burden of selection for software developers when developing service-based systems. To solve this problem, people have proposed various methods to recommend services which are composed into an application. Nevertheless, most of the existing service recommendation methods cannot serve iterative scenarios, i.e., multiple request–response recommending rounds, which frequently occur in real application development. Moreover, they usually fail to utilize full features such as user requirements and service categories, leading to poor performance of service recommendation. To solve the above problems, we propose an iterative framework for service recommendation through multi-model fusion and multi-task learning, called ISRMM. More specifically, we design two models to capture the preferences of applications towards services, through the perspectives of user requirements and history interaction respectively. The output features of the above models are further fused to predict the next service that will be recommended. In addition, we add a tag judgment task to make our framework capable of multi-task learning, through which, the training signal information implied can be used as an inductive bias to improve service recommendation capabilities. Extensive experiments on real datasets show that ISRMM outperforms several state-of-the-art service recommendation methods in iterative service recommendation scenarios.