Section 10-1 to 10-5 Relation Extraction.md

Relation Extraction

10-1 Extracting relations from text

	1. relation extraction’s use
		create new structured knowledged bases
	2. Automated Content Extraction (ACE)
		some kinds of relation combination
	3.Relations databases drawing from Wikipedia
		DBPedia: 1 billion RDF triples, 385 from English Wikipedia
	4.Ontological relations
	5.How to build extractors
		Supervised Learning
		Using patterns

10-2 Using Patterns to Extraction Relations

	1. Hearst’s Patterns for extracting relations
		i.e: X and other Y
	2.Some Richer Relations Using Rules
		Pattern: ORG-LOC, PER-ORG, DRUG-DISEASE
	3.Named entities are not enough for the relations
		some relations are different between same entities

10-3 Supervised Relation Extraction

	1.Supervised Learing
		1)choose relations
		2)choose a set of named entities
		3)choose a representative corpus
		label the named entities in the corpus
		hand-label the relations between these entities
		break into training, development and test
	2.Word features
		M1, M2 ‘s relation:
		M1, M2’s headwords,bigram, type, level(noun)
		NP(noun phrase), VP, PP...

10-4 Semi-Supervised Relation Extraction

	1.relation bootstrapping(Hearst 1992)
	seed
	2.Snowball
	3.distant supervision
		combine bottstrapping with supervised learning
	4.Distantly supervised learning of relation extraction patterns
		1)For each relation i.e. born-in
		2)For each tuple in big database <Elbert, Hubble>
		3)Find sentences in large corpus
		4)Extract frequent features (and later can be used by parsers)

Tip:DBPedia