Towards instance coreference resolution in a multi-ontology environment

With the growing amount of semantic data published on the Web the problem of coreference resolution gains in importance. The linked data initiative provided guidelines for publishing RDF datasets and new datasets are constantly being made available. Such datasets often contain descriptions of the same real-world entities but use different URIs to refer to them. In order to utilize published data on a web scale it is essential to detect such situations and resolve coreferences. In the Semantic Web community initially research effort was primarily concentrated on schema-level ontology alignment and many tools have been developed [1]. With the growing amount of published data instance-level integration issues also started to receive attention recently [2], [4]. These systems abstract from schema-level issues and focus on finding coreferent instances assuming their type and structure to be the same. In our view there is still a gap concerning the study of the complete data integration workflow. On the one hand, schema alignment algorithms do not support the level of granularity necessary for data processing (e.g., applying different settings for individuals of different class). On the other hand, data-level integration tools assume schema-level issues to be resolved and do not consider implications of automated schema alignment. Our system KnoFuss was initially developed to perform integration of automatically extracted annotations structured according to a single common ontology. We extended it to operate in a multi-ontology environment and to utilise schema alignments produced by automatic ontology matching tools. Here we describe the resulting system workflow and first findings obtained during initial tests.