@article{2415,
  author = {Nadia Alboukaey, Ammar Joukhadar},
  title = {Ontology Matching as Regression Problem},
  journal = {Journal of Digital Information Management},
  year = {2018},
  volume = {16},
  number = {1},
  doi = {https://doi.org/10.6025/jdim/2018/16/1/33-42},
  url = {http://dline.info/fpaper/jdim/v16i1/jdimv16i1_4.pdf},
  abstract = {Ontology matching is one of the most important
works to achieve the goal of the semantic web.
To fulfill this task, element-level matching is an indispensable
step to obtain the fundamental alignment. In
element-level matching process, previous work generally
utilizes multiple measures to compute the similarities
among elements, and then combine these similarities
using a weighted sum formula to determine later the semantic
correspondences. The proper selection and combination
of these similarities strongly influence the final
quality of the matching system.
In this paper, we introduce element-level ontology matcher
as machine learning system which utilizes regression algorithms
to automatically find many possible combination
relations of similarity measures. We adopt different
similarity measures to extract learning features, and train
the system on sample data from Benchmark and Conference
tracks OAEI 2015. To match two ontologies, after
training, we measure different similarities for each entity
pair, and predict the overall similarity of each pair using
the learned regression model to get similarity matrix. After
that we extract correspondences by applying naive
descending algorithm on the similarity matrix. These correspondences
are filtered by previously known semantic
techniques to enhance matching precision while preserving
matching recall.
Experimental results, using dataset in conference track
from OAEI 2015, show that our extracted similarity features
are efficient in terms of f-measure evaluation criteria,
and outperform the widely used measures in ontology
matching systems. Moreover, our method for similarity
combination which depends on regression model outperforms
the present combination methods. Besides, in comparison
to the matching systems participated in OAEI 2015 campaign, our matching system showed to be highly
competitive and had a high ranking position.},
}