Matching crowdsourcing There may be several motivations for crowdsourcing ontology matching, i.e., relying on a crowd of workers for establishing alignments [2]. It may be for matching itself or for establishing a reference alignment against which matchers are evaluated. It may also be possible to use crowdsourcing as complement to a matcher, either to filter the finally provided alignment or to punctually provide hints to the matcher during its processing. The ideal way of crowdsourcing ontology matching is to design microtasks (t) around alignment correspondences and to ask workers (w) if they are valid or not.
This will require slightly more work from workers, but they will not require them to choose between alternatives when they do not see any clear correct one. Complement crowdsourcing One further possibility, instead of asking people what could be an answer is to ask them what is definitely not an answer. In this second setting, it may be easier for people to provide meaningful information without needing to commit to one particular answer.
ccw(t) = fv; Gg :(House v Building) ^:(House G Building)
Complement crowdsourcing is logically the complement of disjunctive
crowdsourcing. However, we conjecture that this will make workers adopt a cautious
attitude, discarding only relations that they really think are wrong.
Summary This is related to the consensus between experts [
With a population W of workers, classical crowdsourcing asks if one relation is true or what is the relation between two entities. So, the result of the task cw is a single relation. Disjunctive crowdsourcing asks which relations could hold, hence, dcw R. Similarly, complement crowdsourcing asks which relations do not hold, hence ccw R. We conjecture that: 8w 2 W; ccw(t) fcw(t)g This would have the good feature to provide better opportunity for consensus because: \w2W ccw(t) \w2W dcw(t) \w2W fcwg(t)
It would be an interesting experiment to check if these modalities allow for less conflicts and more accurate alignments. We could test the hypothesis that if it is better to ask users to choose one relation between two entities or to discard nonapplyable relations among all the possible ones.