Incorporating semantics in pattern-based scientific workflow recommender systems: Improving the accuracy of recommendations

Recommender systems are used to enable decision-support. Using them to assist users when designing scientific workflows introduces a number of challenges. These include selecting appropriate components and specifying correct parameter values. Pattern-based workflow recommender systems employ historical usage patterns to generate recommendations. Such systems can intelligently adapt with use. Semantics, on the other hand, can enable recommender systems to intelligently infer new relationships between workflow components. Combining both approaches can help to overcome the drawbacks of each approach and improve the accuracy of the suggestions. To this end, a framework for a hybrid workflow design recommender system is presented in this paper along with the accompanying suggestion generation algorithm. An illustrative example is also presented to demonstrate how the system helps in constructing a workflow. The performance of the framework is compared with an existing pattern-based system using a dataset of neuroimaging workflows. The evaluation results demonstrate that the proposed system outperforms the existing system in a number of different scenarios. The improvement in the performance of the proposed system enhances the usability of the system for users and allows them to more efficiently construct workflows.