TransE Initial Results
Dataset can either be (i) representative of wikidata and have 20% qualifier info (ii) agnostic of real wikidata and have 100% qualifier info.
Created on September 9|Last edited on September 9
Comment
When using 20% qualifier data
Dataset
Our dataset is born out of Fb15k. We have approximately 50k entities in our dataset, which is organized as quints.
Quint looks like
s p o qp qe.
Approximately 20% of the quints/triples in the dataset have actual qualifiers. Rest have s p o __ __, which is representative of Wikidata in general. 288k instances
Runs
solar-donkey-106 is using the quints in training and while predicting.
iconic-vortex-84 uses the triple information from the quints (s p o). Consequentially the training data is slightly less, too.
Run set 1
2
When using a 100% qualifier data
Run set 1
3
WikiPeople
Run set 1
3
Question
57
Add a comment