Evaluating Good Answers in Open Knowledge

The nature of the OpenKnowledge system makes insisting on perfect matching an unfeasibly stringent requirement. If we wish to facilitate the interactions of an unlimited number of disparate peers without prior agreement, it will often be the case that peers can more or less perform a given role, but will find that the role description does not fit perfectly with its own description of its abilities. To make the OpenKnowledge system usable, we need to be able to look for peers that can exactly perform a given role but, if, as will often happen, this search fails, we need to find peers that can approximately perform such roles, and we need some way of estimating how good this approximation is.