Data-driven synthesis of multiple recommendation patterns to create situational Web mashups

Abstract

As a typical situational application, Web mashup reflects and accommodates some key features of Internetware paradigm. Mashup provides a development fashion that integrates data, computation and UI elements from multiple resources into a single Web application, and promises the quick rollout of creating potential new functionalities opportunistically. This paper focuses on the problem of recommending useful suggestions for developing data-driven mashups by synthesis of multiple patterns. We present a rapid and intuitive system called iMashupAdvisor, for aiding mashup development based on a novel automated suggestion mechanism. The key observation guiding the development of iMashupAdvisor is that mashups developed by different users might share some common patterns, for instance, selecting similar mashup components for similar goals, and gluing them in a similar manner. Such patterns could reside in multiple sources, e.g., the data dependency between mashup components, the interaction between users and mashup components, or the collective intelligence from existing applications created and maintained by programmers, etc. iMashupAdvisor leverages the synthesis of these patterns to recommend useful suggestions for a partial mashup, such as the missing components, connections between them, or potentially relevant options, to assist mashup completion. This paper presents the data model and ranking metrics of the synthesis process, and introduces efficient algorithms for the retrieval of recommendations. We also experimentally demonstrate the efficiency of our approach for benefiting the proposed rapid mashup development.