Matcha is a novel ontology matching system under development that aims to tackle holistic ontology matching, complex ontology matching and machine-learning based matching. It builds upon the success of AgreementMakerLight (AML), but is based on an entirely novel architecture to support more complex algorithms. Matcha-DL is an expansion of Matcha for supervised learning settings. Matcha achieved state of the art performance in several OAEI tasks, whereas Matcha-DL ranked first in F-measure in the majority of the Bio-ML tasks.
still under development, some functionalities of AML have not yet been implemented, including translation and alignment repair.
Matcha-DL implements a dense neural network that receives as input the scores generated from running all the matcher algorithms of the Matcha framework. Afterwards, Matcha-DL can also exploit Matcha’s filters for holistic matching to enforce alignment consistency. Currently, Matcha-DL only supports equivalence matching, but an adaptation for subsumption matching is under development. We are also expanding this framework with more complex deep learning based solutions.
No specific adaptations were made to Matcha for the evaluation other than implementing
the MELT [
Matcha-DL supports global matching and local ranking-based metrics. However, it
has a slight disadvantage on the latter since the sparsity in the core Matcha algorithm’s
scores leads to a near binary probability distribution. As such, Matcha-DL is sub-optimal
to optimize metrics like MRR and Hits@k proposed by [
Matcha is still under development and therefore not yet publicly available. A public release will be made once the core development is completed.
Matcha’s OAEI 2022 results are summarized in Table 1, with Matcha and Matcha-DL’s results in the Bio-ML track are reported in Table 2.
The Matcha system is still in its early stages of development and participating in OAEI 2022 was fundamental to highlight challenges and opportunities going forward. We applaud the novel tracks at OAEI 2022 which have shown both the maturity of equivalence-oriented ontology alignment approaches and the opportunites aforded by supervised learning, but also that both ensuring generalizability and tackling tasks such as subsumption and complex matching are still very much open challenges. This work was supported by FCT through the LASIGE Research Unit (UIDB/00408 /2020 and UIDP/00408/2020). It was also partially supported by the KATY project which has received funding from the European Union’s Horizon 2020 research and innovation program under grant agreement No 101017453. Marta Silva was partially funded by FCT through the fellowship 2022.11895.BD.