An Interactive Service Recommendation Method Based On Multiple Interaction And Attention Mechanism

In recent years,software developers have increasingly built a Mashup that can provide specific functions by integrating several Web services.However,the massive increase in Web services has brought severe information overload problems to developers.In this context,service recommendation has emerged as a critical technology that recommends appropriate component services for developers.This paper discusses a common Mashup development scenario: the developer completes the development of a new Mashup in an online session.However,it is hard for most service recommendation methods to handle the cold start problem when developers have not selected any component services in this scenario.Besides,most service recommendation methods cannot promptly update the recommendation result based on their requirements or selection behaviors in the current session.Considering that existing service recommendation methods can hardly work well in the online Mashup development scenario,we propose a multi-round interactive service recommendation approach.Our approach can be divided into two stages.In the first stage,developers have not selected any component services,and the service recommendation system(SRS)analyzes their requirements and returns a list of candidate services,from which developers can select one or more candidate services.If the recommendation result does not fully satisfy the requirements,the recommendation process steps into the second stage,and SRS generates a new service list according to developer requirements as well as the selected component services.In the first stage of the iterative Mashup development,a Mashup to build is essentially an entirely new "user." This paper proposes a novel multiple-interactionsoriented service recommendation method for this cold-start problem.The core of this method is a deep-learning-based hybrid model.Based on the content information of the target Mashup and a candidate service(i.e.,keyword-based developer requirements and service descriptions)as well as the invocation history between neighbor Mashups(i.e.,some existing Mashups similar to the target Mashup)and the service,the model learns the multiple interactions between the Mashup and the service from different aspects and predicts the probability of the Mashup selecting the service.Experiments on real datasets indicate that our method achieves significantly better performance than the other six benchmark methods in this cold-start Mashup development scenario.When the developer has selected one or more component services for the new Mashup,the iterative Mashup development steps into the second stage.This paper proposes an interactive service recommendation framework(i.e.,DLISR)based on selected services to help developers select a next component service.DLISR applies the attention mechanism to measure the impact of different selected services on the next-service selection.According to this framework,this paper proposes two models to leverage content information and historical invocation information to learn the interaction between the target Mashup,selected services,and each candidate service.We then designed a hybrid model called HISR to integrate the above two models.Experiments indicate that HISR outperforms other service recommendation methods to help developers iteratively develop a new Mashup.