Activity Matching with Human Intelligence

Effective matching of activities is the first step toward successful process model matching and search. The problem is nontrivial and has led to a variety of computational similarity metrics and matching approaches, however all still with low performance in terms of precision and recall. In this paper, instead, we study how to leverage on human intelligence to identify matches among activities and show that the problem is not as straightforward as most computational approaches assume. We access human intelligence (i) by crowdsourcing the activity matching problem to generic workers and (ii) by eliciting ground truth matches from experts. The precision and recall we achieve and the qualitative analysis of the results testify huge potential for a human-based activity matching that contemplates disagreement and interpretation.