HSC-Rocket: An interactive dialogue assistant to make agents composing service better through human feedback

Anonymous

Abstract: Facing the current dynamic service environment, fast and efficient service composition has attracted great attention in recent years. Users prefer to express their personal requirements based on natural language, and their real-time feedback could better reflect the effect of service composition to a great extent. Consequently, this paper designs an interactive dialogue assistant, HSC-Rocket, to better provide service composition by considering human feedback. Firstly, we propose a human-computer interaction dynamic service composition algorithm based on reinforcement learning. The design of the reward mechanism considers the quality of service (QoS) and real-time feedback, which can more accurately meet the demands of users. Then, the functional requirements are analyzed through word embedding, to realize the dynamic composition of abstract and concrete services. Furthermore, we utilize the sample enhancement method to alleviate the issue of fewer sample data in the initial stage of user interaction, which improves the robustness of our system. Accordingly, we have implemented the HSC-Rocket prototype, which allows users to obtain multi-domain dialogue requirements. Extensive experiments on the RapidAPI dataset have demonstrated the superiority and effectiveness of the HSC-Rocket.