In many applications, the key question is how much accurate knowledge can be extracted. Knowledge base population (KBP) is broken down to two subtasks:

  1. Entity linking (discussed already)
  2. Slot filling

Differences between slot filling and sentence relation extraction

  1. Extracting relations hold between entities rather than spans of text

  2. Performance of slot filling is evaluated at the type level (on entity pairs) rather than on individual sentences

  3. KBP system often starts with information retrieval phase (extracting relevant passages)

  4. There will be multiple passages of text for every entity pairs. Slot filling system needs to aggregate the evidence to predict a single relation type (or set of relations)

  5. Labelled data is in the form of pairs of related entities rather than annotated passages of text. This could be an issue as two entities may be linked by several relations



Data Scientist

Leave a Reply