A System-of-Systems (SoS) is a set of independent information systems that must communicate with each other towards providing a speci c service. Therefore, e ectively integrating these systems is demanding. Considering that each system is conceptually described by a unique ontology, the conceptual support for the whole SoS demands the alignment of all ontologies, deriving a network of ontologies. Existing ontology matching techniques may be used for the task; however, due to the recently increasing size of the ontologies and the potential number of ontologies being aligned, current approaches may su er from scalability and performance issues. In this paper, we introduce an approach to reduce the number of potential correspondences, therefore optimizing the process of creating a network of ontologies. A preliminary experiment was conducted, showing the potential of the proposed approach.
A System-of-Systems (SoS) is de ned as a set of independent information systems
(IS), providing functionalities derived from the interoperability among them [
Traditional solutions for ontology matching may be applied for solving the
internetwork matching problem, either using a pairwise or a holistic strategy
[
This work proposes an optimized approach for the internetwork matching
challenge [
This work is organized as follows: Section 2 de nes the internetwork matching problem. Our proposed approach is detailed in Section 3. Preliminary evaluation results are in Section 4. Finally, section 5 concludes and points to future work. 2
A network of ontologies is formally de ned as =< ; >, where is a nite
set of ontologies and (O; O0) is a set of alignments between pairs of ontologies
belonging to [
One of the approaches for matching networks of ontologies is pairwise. Given a set of networks of ontologies, the pairwise internetwork matching sequentially computes the alignment of each pair of ontologies from each pair of networks from this set. For example, given two networks of ontologies =< ; > and 0 =< 0; 0 >, in which = fO1; O2g and 0 = fO3g, the pairwise internetwork matching is obtained by computing (((O1 O2) [ (O1 O3)) [ (O2 O3)). That is, the pairwise internetwork matching approach computes all matchings between all pairs of ontologies inside each network that is being aligned.
Networks frequently have isomorphisms and trivial alignments that may cause the pairwise approach to nd the same alignments more than once, thus requiring an additional step to merge the resulting matches at the end. In the case of isomorphisms, identical correspondences between same entities may be generated (for instance, in Figure 1 a matcher tool may nd A1;10 = f< O1:a1; O10:a01; => ; < O1:b1; O10:b01; =>g). The case of a trivial alignment occurs when a group of entities, that was previously aligned in a network, appears in another network. For instance, a pairwise matcher that receives the networks and 0 and has previously computed the intra-network alignments A1;2 =< O1:b1; O2:d2; v> and A10;20 =< O10:b01; O20:d02; v> will work unnecessarily to produce A1;20 =< O1:b1; O20:d02; v> and A10;2 =< O10:b01; O2:d2; v>, which are trivial alignments.
The main weakness of pairwise approach is the number of comparisons needed to compute all alignments and the lack of ability to handle isomorphisms and intra-network alignments. Given the limitations discussed in Section 2, we propose SubInterNM, a new subsumed approach for the Internetwork matching problem. SubInterNM avoids unnecessary computation by identifying and reusing trivial and subsumed alignments already computed in the networks of ontologies that are being aligned. In such cases, as the intersection among the networks becomes larger, the set of evaluated correspondences during the alignment process tends to get smaller.
For instance, consider the example depicted in Figure 1. The networks and 0 were aligned using an internetwork matcher. Since O2 and O20 are identical, they do not need to be exhaustively compared. Also, both pairs of ontologies O1 and O10 and O3 and O30 share some subset of entities, thus common parts could be eliminated to compute the nal network of ontologies resulted from the internetwork matching problem scenario.
Casanova et al. [
Our proposed approach SubInterNM uses a combined set of lightweight
operations from [
Our proposed approach was evaluated in a preliminary experiment using ontologies from the OAEI conference dataset. We experimented 5 distinct scenarios, and in each of them we speci ed two networks of ontologies to be aligned. The experiments were de ned by increasing the number of ontologies in the network, as well as varying the ontologies and the number of common ontologies between the networks, in order to assess how well our approach would handle the existence of trivial and subsumed alignments: = fconference, cmtg and 0 = fcmt, sigkddg; = fconference, cmt, ekawg and 0 = fcmt, sigkddg; = fconference, cmt, ekawg and 0 = fcmt, sigkdd, conferenceg; = fconference, cmt, dblp, ekawg and 0 = fcmt, sigkdd, conferenceg; = fconference, cmt, edas, ekawg and 0 = fcmt, sigkdd, conferenceg.
In order to compare our proposed SubInterNM approach against the pairwise
internetwork matching approach, we implemented the pairwise approach using
the existing matching system ALIN [
Table 1 shows the number of pairs analyzed (i.e., comparisons) by pairwise and subsumed approach to compute alignments in each experiment. It is possible to verify that SubInterNM reduces the number of comparisons needed to the matching process by at least 24%, therefore representing a successful way to deal with large networks in the internetwork matching problem.
Future work will try to reduce even more the number of comparisons with the
detection of intra-network alignments. All the data gathered in the experiment is
available at (https://bit.ly/2M3jIYS).
This work addressed the problem of internetwork ontology matching, a natural
evolution of the classical ontology matching problem for highly interconnected
scenarios of Systems of Systems, which are of increasing popularity and relevance.
We proposed SubInterNM, an approach for internetwork ontology matching
that optimizes the required computation of correspondences by identifying and
reusing trivial and subsumed alignments. Preliminary evaluation results showed
the potential of the approach and opportunities for improvement, in scenarios
using lightweight ontologies. The computation of subsumed networks posed an
overhead computation since it is a well-known NP-Hard problem [
In future work, we expect to improve our implementation, including parallel programming and infrastructure. We also plan to move forward exploring trivial alignments in newer versions of the tool.