Negative sampling

[TGDivy]

Hi,
In the dataset file for the baseline,
line 128.
knowledge_key = "{}__{}__{}".format(knowledge["domain"], knowledge["entity_id"], knowledge["doc_id"]) # find snippets with same entity as candidates prefix = "{}__{}".format(knowledge["domain"], knowledge["entity_id"])
I think the code might be intended to select knowledge snippet from a specific document. I.e. if the label was hotel__1__15, then to select all hotel__1__. However, because of the code it also ends up selecting documents like hotel__11__, hotel__12__*, etc. This could potentially be fixed.