On-the-fly Learning-based Search for QoS-aware Web Service Composition

The Web Service Composition (WSC) is a prominent technique to help software developers to easily build applications on top of the Service Oriented Architecture. Given a set of web services and a user request, the aim of the WSC problem is to construct an optimal composite web service to satisfy the request. In this paper, in particular, we study the Quality of Services (QoS)-aware WSC problem to optimize the quality criteria of the composite service, e.g., throughput, availability, response time, capacity and accuracy. However, since the QoS-aware WSC problem corresponds to a global optimization problem, it is a hard problem to solve the problem for large scale instances. To resolve this challenge, we propose a novel solution using on-the-fly learning-based search. Our technique employs the Learning Depth-First Search (LDFS) as an underlying search algorithm, which performs iterated depth-first searches enhanced with learning. Moreover, the on-the-fly searching technique partially constructs a search graph only when the part of the graph is required. We empirically show, with a number of examples, that the proposed on-the-fly technique is able to find the optimal composite services much earlier than our previous work, the off-line LDFS method.