search

sefeoglu / ODS_project_student

A new ontology matcher.
GNU General Public License v3.0
0 stars 2 forks source link

Ask Francis' repo #2

Closed sefeoglu closed 10 months ago

sefeoglu commented 12 months ago

Contact Francis about his source codes at his paper.

sefeoglu commented 11 months ago

Tree_walk from Francis tree_walk

Dear Tazio,

First of all, I am sorry for my late response. Since last year I have continued my work and the SEBMatcher project has been transformed into SORBET. Which is an Ontology embedding method that can be used for both Ontology alignment and subsumption prediction. Our system in the OAEI 2023, SORBETMatcher, has also achieved better results than SEBMatcher while also being an unsupervised system.

Having said that, you might want to check out SORBET, it also contains a better version of SEBMatcher's Random Walk algorithm which lead to better performances.

In any case, the code for the random walk, can be found at: https://github.com/Lama-West/SORBET_ISWC23/blob/main/src/batch_loaders/random_walk.py

The Random walks are used in SEBMatcher's batch loaders here: https://github.com/Lama-West/SORBET_ISWC23/blob/main/src/batch_loaders/MLM_batch_loader.py#35 and here: https://github.com/Lama-West/SORBET_ISWC23/blob/main/src/batch_loaders/pair_alignment_batch_loader.py#L24

The Tree Walk may be better explained in the SORBET paper (see .png attached).

As for hyperparameters, they changed a lot since the start of the project. Usually having a max length smaller than 6 and a number of branch randomized between 0 and 3/4/5 is good.

I hope this helps you, Feel free to ask for anything.

Francis Gosselin