Interesting that enforcing an ontology seemed to result in reduced performance. (https://youtu.be/hLDKW3-OKUA) I guess developing an ontology based on the extracted graph will be an important area for future work
Makes sense. If you’re too specific early on, you miss everything that you didn’t expect. Much better to be open first, and then do the filtering afterwards, based on what you observe.
Repo in paper - https://github.com/stair-lab/kg-gen/
nice
Interesting that enforcing an ontology seemed to result in reduced performance. (https://youtu.be/hLDKW3-OKUA) I guess developing an ontology based on the extracted graph will be an important area for future work
Makes sense. If you’re too specific early on, you miss everything that you didn’t expect. Much better to be open first, and then do the filtering afterwards, based on what you observe.