Research On Anchor Based Flooding Mapping Algorithm For Large Scale Ontology

It is a great challenge to match two massive ontologies efficiently; the existing strategies for large-scale ontology are mostly based on partition and then mapping in the small segmented blocks. However, the process of partitioning and finding related segmented blocks is time-consuming, and some sementic information will be lost inavoidably because of partition. Therefore, a highly scalable algorithm will be researched in this thesis.Firstly, background of the research is introduced simply with a general summary of current methods on matching large-scale ontology.Secondly, anchor auto-search module is designed using conceptual type technology to gain anchors (a pair of "look-alike" concepts) quickly. Then, flood mapping algorithm with collision avoidance and unilateral flag strategy is proposed, which starts off with an anchor, gradually exploring new anchors by comparing concepts around the anchors taking advantage of locality of reference in the RDF directed graph and checking old anchor with the new gained anchors, the process is repeated after new anchor is gained until there is no new anchor found.Thirdly, in each flood operation, neighbor concepts around the anchor from the two ontologies are candidate mapping sets, based on which, a vector space is constructed and concepts are vectorized using virtual document technology, then, similarity between concepts is gained through computing cosine value of the angle of vectors.Finally, the experiment shows the features:1. Anchor auto-search is supported;2. The candidate sets for comparison are restricted around the anchors all along, hence, time complexity is greatly reduced.3.1:n mappings can be discovered;4. Most misaligned anchors gained from auto-search module can be removed by checking their structural similarity, hence, precision is greatly improved;5. Flood operation with collision avoidance can improve both mapping efficiency and quality.