A major problem with integrating knowledge bases arises from the differing taxonomies underlying individual knowledge bases. Two nodes, one from each of a pair of taxonomies, may be related in one of four ways: they may be equivalent; the first may be a generalization, or a specialization, of the second; or none of these three cases applies, in which case the authors call the relationship “other.”

This paper describes a method for taxonomy integration that uses crowdsourcing to match two taxonomies. An important feature of the authors’ approach is that they take into account the costs involved in invoking crowdsourcing. Given two taxonomies, T1 and T2, the problem becomes one of selecting k nodes from T2 to which to apply the crowdsourcing classification. The number k is determined by the cost of the crowdsourcing.

In the matching of each node, it is necessary to supply some context information from the two taxonomies. Selecting the k queries that maximize the utility of the queries is shown to be an NP hard problem. Another approach that the authors use is one that they call blocking-based instance matching in which similar entities are clustered and the matching is conducted only within clusters. The authors describe the application of their ideas to the alignment of two knowledge bases, YAGO (http://www.mpi-inf.mpg.de/departments/databases-and-information-systems/research/yago-naga/yago/) and DBpedia (http://wiki.dbpedia.org/). Both of these have several million instances and thousands of classes.

The paper presents an interesting and potentially valuable approach to a significant problem. The ideas are illustrated with illuminating simple examples.