RinsMatch: a suggestion-based instance matching system in RDF Graphs

Introduction. In this paper, we present RinsMatch (RDF Instance Match), a suggestion-based instance matching tool for RDF graphs. RinsMatch utilizes a graph node similarity algorithm and returns to the user the subject node pairs that have similarities higher than a defined threshold. If the user approves the matching of a node pair, the nodes are merged. Then more instance matching candidate pairs are generated and presented to the user based on the common predicates and neighbors of the already matched nodes. RinsMatch then reruns the similarity algorithm with the merged RDF node pairs. This process continues until there is no more feedback from the user and the similarity algorithm suggests no new matching candidate pairs. In our previous study [1], we proposed an algorithm for computation of entity similarities of an RDF graph using graph locality, neighborhood similarity, and the Jaccard measure. In the current study we use the proposed RDF entities similarity algorithm for pairing entities which may be merged if approved by the user. We make a similar assumption like the similarity flooding (SF) algorithm proposed in [2], that elements of two graphs are similar when their adjacent elements are similar. Comparing to SF, our technique requires more user interactions and more iterations for computation of entity similarity, but each time the similarity algorithm runs, it produces more accurate results assuming the user provided accurate feedback. Also, merging the RDF nodes reduces the size of the input data graph that the algorithm operates on, yielding less complexity each time.