Ontology-Driven Mashup Auto-Completion on a Data API Network

Abstract

The building of data mashups is complicated and error-prone, because this process requires not only finding suitable APIs but also combining them in an appropriate way to get the desired result. This paper describes an ontology-driven mashup auto-completion approach for a data API network to facilitate this task. First, a microformats-based ontology was defined to describe the attributes and activities of the data APIs. A semantic Bayesian network (sBN) and a semantic graph template were used for the link prediction on the Semantic Web and to construct a data API network denoted as N_p. The performance is improved by a semi-supervised learning method which uses both labeled and unlabeled data. Then, this network is used to build an ontology-driven mashup auto-completion system to help users build mashups by providing three kinds of recommendations. Tests demonstrate that the approach has a precision_p of about 80%, recall_p of about 60%, and F_0.5 of about 70% for predicting links between APIs. Compared with the API network N_e composed of existing links on the current Web, N_p contains more links including those that should but do not exist. The ontology-driven mashup auto-completion system gives a much better recall_r and discounted cumulative gain (DCG) on N_p than on N_e. The tests suggest that this approach gives users more creativity by constructing the API network through predicting mashup APIs rather than using only existing links on the Web.