Flexible service discovery based on multiple matching algorithms

Abstract

Traditional web service discovery approaches rely on logic or non-logic matching techniques. In general, logic approaches can achieve satisfactory precision levels, but they result in modest recall scores. In contrast, non-logic approaches may ensure more balanced scores in terms of recall and precision, but they need additional aggregation schemes or optimisation methods. To improve the discovery performance, we need to combine multiple matching algorithms and fuse their results into a single ranked list of services. This combination must avoid the well-known side effects of fusion, such as overfitting or noise sensitivity. To tackle the service-discovery issue, we propose a solution based on two key ideas: first, we propose a majority voting model based on the 'Condorcet' paradigm to fuse a set of individual ranked lists (provided by the matching functions). Second, we leverage a probabilistic extension of the dominance relationship to ensure comparison between the services. The experimental evaluations indicate the proposed solution, 'probabilistic Condorcet', outperforms all individual matching functions, as well as many concurrent fusion algorithms.