TAC-KBP 2009 Track

A TAC-KBP 2009 Track is a TAC-KBP track within TAC 2009.

• AKA: TAC 2009 Knowledge Base Population Track.
• Context:
  • It can result in a TAC-KBP 2009 Task.
  • It can result in a TAC-KBP 2009 Papers (presented at the TAC 2009 workshop).
• Example(s):
  • Knowledge Base Population (KBP).
• Counter-Example(s):
  • TAC-RTE 2009.
  • TAC Summarization 2009.
  • TAC-KBP 2014 Track.
  • TAC-KBP 2011 Track.
  • TAC-KBP 2010 Track.
• See: TAC Track, TAC Workshop.

---

---

References

2009

• http://apl.jhu.edu/~paulmac/kbp.html
  • Overview: Question Answering and Information Extraction have been studied over the past decade; however evaluation has generally been limited to isolated targets or small scopes (i.e., single documents). At TAC 2009 the Knowledge Base Population (KBP) Track will explore extraction of information about entities with reference to an external knowledge source. Using basic schema for persons, organizations, and locations, nodes in an ontology must be created and populated using unstructured information found in text. A collection of Wikipedia Infoboxes will serve as a rudimentary initial knowledge representation.

• (McNamee et al., 2009) ⇒ Paul McNamee, Heather Simpson, and Hoa Trang Dang. (2009). “Overview of the TAC 2009 Knowledge Base Population Track. "
  • IE & QA technologies have been studied in isolation
    • Not focused on discovery of information for inclusion in an existing knowledge base
    • No consideration of novelty, contradiction
  • Issues when filling in a KB
    • Accurate extraction of facts
    • Global resolution of entities
    • Maintaining provenance of asserted facts
    • Avoiding contradiction / detection of novel information
    • Temporal qualification of assertions
    • Leveraging existing KB to assist with extraction
    • Scalability
  • Comparison to ACE & TREC-QA
    • Corpus vs. document focus
    • ACE: component tasks (NER, relation extraction) for a set of isolated documents
    • KBP: learn facts from a corpus. Repetition not very important. Asserting wrong information is bad.
  • Context
    • In KBP, there is a reference knowledge base, so avoiding redundancy and detecting contradiction are important
    • In KBP slots are fixed and targets change. In TREC QA, the targets dictated which questions were asked.
  • Knowing when you don’t know
    • TREC QA had a small percentage of NIL questions (4-10%)

---