Abstract
The consumption of APIs, such as Enterprise Services (ESs) in an enterprise
Service-Oriented Architecture (eSOA), has largely been a task for experienced developers.
With the rapidly growing number of such (Web) APIs, users with little
or no experience in a given API, e.g., casual developers creating mashups, face the
problem of trying to find relevant API operations. However, building an effective,
easy-to-use search has been a challenge: Information Retrieval (IR) methods struggle
with the brevity of available text in API descriptions, whereas semantic search
technologies require available domain ontologies and queries formulated in formal
languages. In the following, we focus on ESs: enterprise-class Web Services, providing
access to enterprise applications, e.g., Customer Relationship Management
(CRM). ESs are commonly developed using service design methodologies, guarded
by SOA Governance, to manage large sets of ESs. In this work, we describe an
end-to-end approach to facilitate the management and search of ESs on the basis of
such knowledge. First, we provide a formal definition of a service design knowledge
base to represent entities and their relationships from service design methodologies.
We further describe an entity matching approach to automatically index large sets
of ESs with entities from such a knowledge base. Second, we present an approach to
facilitate the typically manual effort of developing knowledge bases. Due to the limitations
of existing techniques to automatically amend knowledge bases, we propose a
novel approach based on clustering, complemented with various filtering and ranking
techniques to identify new entities from a set of existing ES operation names. Third,
motivated by the search behavior of users, we propose an iterative keyword search
based on entities from a service design knowledge base. We hereby describe a novel
ranking technique based on different ranking components related to service design
knowledge and meta-data derived from the service repository infrastructure. Finally,
we implemented and evaluated prototypes for the entity matching, knowledge base
amendment and keyword search using a knowledge base with more than 1500 ESs
from SAP. The results are highly encouraging and show significant improvements
over state of the art entity matching, clustering and IR techniques.