Replies: 1 comment 1 reply
-
Weighting triples is an open research question. You can be creative by writing your own interaction function, where you can arbitrarily modify the way that a triple is scored. You can learn more about doing this here: https://pykeen.readthedocs.io/en/stable/byo/interaction.html. Alternatively, you can write your own loss function, which is responsible for aggregating scores for triples. One (proposed) example for doing this is in #1142, however, you won't need such a complicated solution for a specific use case as the author of this PR is proposing, since they want it fully baked into the framework. |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Tried to do so using custom interactions/models but didn't manage to make much progress. I want to create a model where e.g. for relationship1 and relationship2, I can associate a different edge weight (giving more importance to certain relationships over others, thus affecting the similarity produced), and then proceed to train a TransE KGE model. Any help would be appreciated!
Beta Was this translation helpful? Give feedback.
All reactions