WaSABi’2013
Adaptive Semantic Publishing
Georgi Georgiev, Borislav Popov, Petya Osenova & Marin Dimitrov
Contents
• Adaptive Semantic Publishing
• Software and Information Architecture
• Semantic Annotation Process
• Personalised Content Delivery
#2 Adaptive Semantic Publishing (WaSABi’2013) Oct 2013
Adaptive Semantic Publishing
• High-quality content
– Recommened related materials
– Improved reuse
• Dynamic products
– Semantic enrichment
– Metadata-driven web content
– Dynamic re-purposing
• Personalised content delivery
– Adaptive content streams
– Highly relevant information
#3 Adaptive Semantic Publishing (WaSABi’2013) Oct 2013
Success Stories
• BBC Sports (UK)
– 2010 World Cup & 2012 Olympics
• Press Association (UK)
– Semantic enrichment of assets
• EuroMoney (UK)
– Finance & economics
• NDP (NL)
– Semantic publishing, 3rd party apps
• Publicis (DE)
– Content recommendation
#4 Adaptive Semantic Publishing (WaSABi’2013) Oct 2013
BBC’s Dynamic Semantic Publishing
#5 Adaptive Semantic Publishing (WaSABi’2013) Oct 2013
Adaptive Semantic Publishing for NDP
#6 Adaptive Semantic Publishing (WaSABi’2013) Oct 2013
XML Content Store
Generalised Semantic Publishing Architecture
#7 Adaptive Semantic Publishing (WaSABi’2013) Oct 2013
Information Architecture
• Subset of LOD
– Dbpedia & Freebase – persons & organisations
– Geonames – important locations
– Tailored for the specific use case needs
– 2.5M+ entities
• LOD ontologies mapped through PROTON
– Upper-level ontology; 250 classes & 100 properties
– Unified ontology interface
• Domain specific ontologies
#8 Adaptive Semantic Publishing (WaSABi’2013) Oct 2013
Annotation Guidelines & Corpus Creation
• Functional requirements
– Informal discussion with customer’s domain experts
– Sample documents <> sample (manual) annotations
– Agreement on annotation types
• Annotation guidelines
– Rules, examples, ambiguous cases, quality checks for annotators
• Manual annotation
– Based on the annotation guidelines
– Multiple annotators + super-annotator
• Initial corpus for evaluation #9 Adaptive Semantic Publishing (WaSABi’2013) Oct 2013
Iterative Improvement of the Semantic Annotation
#10 Adaptive Semantic Publishing (WaSABi’2013) Oct 2013
Curation
• Instance disambiguation
– “Washington” >> Politician, US State, US City, …
• Tagging
– Verify automated annotations
– Fix mis-tagged entities if necesary
• Topic / keyword correction
#11 Adaptive Semantic Publishing (WaSABi’2013) Oct 2013
Curation
#12 Adaptive Semantic Publishing (WaSABi’2013) Oct 2013
Sample Evaluation Corpus
• 250 documents, 4.5K average document size
• 6,000+ manually annotated entities
• 2.35 candidates for disambiguation per entity
– … but up to 100 candidates in some cases!
• F1 (Org) = 0.82; F1 (Per) = 0.90; F1 (Loc) = 0.92
#13 Adaptive Semantic Publishing (WaSABi’2013) Oct 2013
Personalised Content Delivery
#14 Adaptive Semantic Publishing (WaSABi’2013) Oct 2013
Personalised Content Delivery
#15 Adaptive Semantic Publishing (WaSABi’2013) Oct 2013
Personalised Content Delivery
#16 Adaptive Semantic Publishing (WaSABi’2013) Oct 2013
Q & A
Thank you!
#17 Adaptive Semantic Publishing (WaSABi’2013) Oct 2013