Unsupervised Ontology Induction From Text - Microsoft Research
Unsupervised Ontology Induction From Text - Microsoft Research Unsupervised Ontology Induction From Text - Microsoft Research
Probabilistic Model for USP • Joint probability distribution over input dependency trees and their semantic parses • Use Markov logic • A Markov Logic Network (MLN) is a set of pairs (F i , w i ) where • F i is a formula in higher-order logic • w i is a real number 1 P ( x ) e x p w N ( x ) i i Z i Number of true groundings of F i
Unsupervised Semantic Parsing • Exponential prior on number of parameters • Cluster mixtures: InClust(e,+c) ^ HasValue(e,+v) Object/Event Cluster: INDUCE induces 0.1 enhances 0.4 … Property Cluster: INDUCER nsubj 0.5 IL-4 0.2 None 0.1 agent 0.4 IL-8 0.1 One 0.8 … … … … 21
- Page 1 and 2: Unsupervised Ontology Induction Fro
- Page 5 and 6: Extracting Knowledge From Text …
- Page 7 and 8: Unsupervised Learning for Knowledge
- Page 9 and 10: Ontology Learning • Attracts incr
- Page 11 and 12: Outline • Background: USP • Uns
- Page 13 and 14: Challenge: Same Meaning, Many Varia
- Page 15 and 16: Unsupervised Semantic Parsing • U
- Page 17 and 18: Unsupervised Semantic Parsing • U
- Page 19: Unsupervised Semantic Parsing • U
- Page 23 and 24: Learning: Hill-Climb Likelihood Ini
- Page 25 and 26: OntoUSP • USP + Hierarchical clus
- Page 27 and 28: Experiments • Evaluate on an end
- Page 29 and 30: Induced Ontology (Partial)
- Page 31: Conclusion • OntoUSP: Unsupervise
<strong>Unsupervised</strong> Semantic Parsing<br />
• Exponential prior on number of parameters<br />
• Cluster mixtures:<br />
InClust(e,+c) ^ HasValue(e,+v)<br />
Object/Event Cluster: INDUCE<br />
induces 0.1<br />
enhances 0.4<br />
…<br />
Property Cluster: INDUCER<br />
nsubj 0.5 IL-4 0.2 None 0.1<br />
agent 0.4 IL-8 0.1 One 0.8<br />
…<br />
…<br />
…<br />
…<br />
21