Abstract
There are a lot of data sources that can be used to make a reasonable decision in the diversity of fields. But the linking of data from these sources into a solid view must come first. To represent linked data RDF language is used. Unfortunately, there are lots of those who publish data in non-RDF formats. Obviously, considering current data volumes integration task cannot be done in analytics’ heads. A number of instruments to link heterogeneous data were proposed. The most flexible ones are based on special mapping languages. However, these solutions lack scaling and may not be effective for processing large data. In the presented paper, an approach for a possible solution is proposed. The idea of the approach is to abstract data linking process model separating it from a mapping as well as from a code implementation. The model is automatically generated based on a mapping of a set of source ontology elements into a set of domain ontology elements. The model of a process becomes available for separate manipulations relatively independent of other activities. The model can be used to optimize or to custom the linking process or can be treated as a specification for different implementations. As it based on the data-flow concepts it can be naturally translated into contemporary highperformance computation concepts. In other words, it is demonstrated how a process model can be implemented on one of such platforms. The obtained results make it reasonable to continue investigating and developing the proposed approach.