Nordic health data metadata
0 likes95 views
This document discusses using metadata and knowledge graphs to better organize health data and make it more findable. It explains how knowledge graphs work by connecting entities and their relationships, and how this can help match user search intent to the meaning of data. The document also discusses challenges in organizing diverse data sources and standards, and how semantic annotation and knowledge graphs can help integrate different data types and make them interoperable.
Nordic health data metadata
- 1. Start Making Sense Be F.A.I.R Nordic HealthData – how do we utilize metadata in health practice?
- 2. INSTITUTIONEN FÖR TILLÄMPAD INFORMATIONSTEKNOLOGI | www.ait.gu.seINSTITUTIONEN FÖR TILLÄMPAD INFORMATIONSTEKNOLOGI | www.ait.gu.se Moore humor anno 1965 Reality today 2020 http://orangecone.com/archives/2010/04/smart_things_ch_2.html
- 3. INSTITUTIONEN FÖR TILLÄMPAD INFORMATIONSTEKNOLOGI | www.ait.gu.seINSTITUTIONEN FÖR TILLÄMPAD INFORMATIONSTEKNOLOGI | www.ait.gu.se It was 20+ years ago today… Dr Pål Lindström Pål, founded: LocusMedicus Int. • A Healthcare Community of Practice • Knowledge Networking automated and augmented by AI
- 4. Challenges
- 5. Landfill
- 8. DATA
- 9. FAIR
- 10. Organising
- 11. Organising
- 12. Decipher
- 13. Lingua Franca 𝑥 + 𝑎 $ = & '() $ 𝑛 𝑘 𝑥' 𝑎$,' 𝑓 𝑥 = 𝑎) + & $(. / 𝑎$ cos 𝑛𝜋𝑥 𝐿 + 𝑏$ sin 𝑛𝜋𝑥 𝐿 𝑎8 + 𝑏8 = 𝑐8 1 + 𝑥 $ = 1 + 𝑛𝑥 1! + 𝑛 𝑛 − 1 𝑥8 2! + ⋯ 𝑥 = −𝑏 ± 𝑏8 − 4𝑎𝑐 2𝑎 sin 𝛼 ± sin 𝛽 = 2 sin 1 2 𝛼 ± 𝛽 cos 1 2 𝛼 ∓ 𝛽 𝐸 = 𝑀𝐶8
- 15. Organising Principles http://flic.kr/p/kRqh42 SKOS DC W3C schema.org Linked Data RDF HL7/FHIR ISA2 MeSH UMLS SnoMed CT IDC11
- 16. INFORMATION ARCHITECTURE AS AN ORGANIZING DISCIPLINE the discipline of organising 1. What Is Being Organised? 2. Why Is It Being Organised? 3. How Much Is It Being Organised? 4. When Is It Being Organised? 5. Who (or What) is Organising It? 6. Where is it Organised? 16 Robert Glushko
- 17. INFORMATION ARCHITECTURE AS AN ORGANIZING DISCIPLINE ● Libraries, markets, museums, zoos, vineyards ● Different types of data and documents ● Personal information and artifacts ● People We Organise 17 Robert Glushko
- 18. INFORMATION ARCHITECTURE AS AN ORGANIZING DISCIPLINE An “Organising System” Robert Glushko 18
- 19. INFORMATION ARCHITECTURE AS AN ORGANIZING DISCIPLINE Organising Books By Content Photo by Jeffrey Beall (http://www.flickr.com/photos/denverjeffrey/304220561) Creative Commons CC BY-ND 2.0 Robert Glushko 19
- 20. INFORMATION ARCHITECTURE AS AN ORGANIZING DISCIPLINE Recall: The “Organising System” A collection of resources intentionally arranged to enable some set of interactions Robert Glushko 20
- 21. INFORMATION ARCHITECTURE AS AN ORGANIZING DISCIPLINE “Information Architecture is designing an abstract and effective organisation of information and then exposing that organisation to facilitate navigation and information use” Intentionalarrangement Interactionsupport Robert Glushko 21 Defining “Information Architecture” as an Organising Discipline
- 22. Standing on the shoulders of Giants No AI without IA Newton ShannonBayes TuringAl Kindi Ada Wolfram
- 23. INFORMATION ARCHITECTURE AS AN ORGANIZING DISCIPLINE The Document Type Spectrum – A Continuum Between Documents and Data Robert Glushko 23
- 24. Tools for a smart platform
- 25. Data Findability: How Knowledge Graphs can support FAIR data
- 26. Google works for us as consumers, yes? Weekend search
- 27. Frustrated by search at work? Handy-patient summary? Latest research? Applicable guidelines?
- 28. Is your data designed to be found/understood? Metadata: what is it? what’s inside it? what are the ingredients? Context: how old is it? when and where was it made? what is needed to access the contents? Structure: does it need to be cooked/processed? how long does it need to be cooked for? Data connections: will it be like anything I’ve had before? will it go with the other items I have? will it be something others like? Data meaning / comprehension (aka Findability)
- 29. Google & its Knowledge Graph
- 30. Bridging the gap: Data meaning v User intent Situation for most organisations Google Google’s Knowledge Graph 2010 (based on Semantic Web-based technologies, incl. RDF)
- 31. A Knowledge Graph: an imagined representation Nodes = entities, concepts, phrases Lines/edges = relationships KG’s underlying tech = Semantic Web-based Technologies - all Worldwide Web Consortium/W3C standards Base model = RDF (Resource Description Framework) RDF triples /predicates Subject Object predicate Value predicate Patient X Loss of energy hasSymptom 38.0°ChasTemperature Every concept and relationship has its own identifier (URI) – less ambiguity
- 32. PageRank – known Webpage connections Pre 2010 Keyword matching Page Rank 50% space given to organic search results
- 33. Connected Data – known relationships Match data meaning with query intent Findable in FAIR + Context awareness (Location, Time of day, Device etc) + UX becomes more intuitive to reflect possible query/intent Search Results and Discovery Information Boxes (auto publishing) Q & As Searches related to “your query” 10% space given to organic search results Google’s Knowledge Graph, 2010 Encoded knowledge Machine readable Human readable Entity extraction Known entities + known relationships with other entities (connected data) People Organisations Places Events Products Services meaning + aboutness
- 34. known entities: Watch (product), sold where, by whom, price etc Q & A Information box Online sellers Organic search results Ads Retail sellers Related searches Location, Device, Time of Day
- 35. known entities: Watch (product), sold where, by whom, price etc Q & A Information box Online sellers Organic search results Ads Retail sellers Related searches Location, Device, Time of Day 30% of websites now help Google with the meaning of their websites by adding (RDF) Schema.org and microformats Includes for populating: Q&As, Info Box, rich snippets etc
- 36. known entities: Virus, infectious agent, diseases caused, related symptoms no Ads Request for feedback Top stories Help & information Safety tips Organic search results Trusted sources
- 37. Organisations can use KGs to make data more findable too (web-base applications)
- 38. Make data smarter by Semantic Annotation Tabular data Metadata “B” Info model Z RDBMs Metadata ”A” Info model Y Archived dataDocuments RDF 1. Data connectors – different data sources and formats 2. NLP (NER) & ML: processing & enrichment pipeline 3. Standardized metadata 4. Data semantically annotated (consistently) with concepts + relationships from the KG 5. Semantic layer over all data sources: Interoperable (FAIR) Elastic KG
- 39. Make search smarter: analyse user behaviour Tabular data Metadata “B” Info model Z RDBMs Metadata ”A” Info model Y PDFsDocuments RDF 1. Search & click log analysis 2. ML: Learning to Rank 3. Detect new concepts for the KG Kibana
- 40. How do you make a Knowledge Graph?
- 41. 2-D elements of a 3-D Knowledge Graph Polyhierarchical Taxonomy / Thesaurus Ontology Main domain class relationships “Business rules” Information model 1. Synonyms, acronyms, abbreviations 2. Language from different perspectives (not just keyword match) 3. Use standard terminologies /pick&mix 4. Different spoken languages 5. Cultural accessibility & interoperability (FAIR) 6. UX typing = autocomplete 7. UX navigation - polyhierarchical hasSymptom hasSymptom Each concept and relationship has a unique URI – helps prevent ambiguity when searching: Cold (common cold) Cold (temperature, feeling)
- 42. Semantic Annotation: Content v Datasets • Content (unstructured text) holds more concepts with which to refer to the knowledge graph • Exploit 80% of unstructured text in EHRs • Connect & index various sources through one search interface – see the latest research and guidelines for the treatment of a disease or condition • Patient summary • Datasets/Databases have fewer concepts, often unclear field names (reference or shortened text) • Harder to find data from keyword matching alone • Knowledge graphs can add a higher level meaning to datasets, if not just the metadata is indexed • E.g. “Hypertension” – all datasets with a systolic blood pressure over 160mmHg
- 43. Different types of metadata Concrete Abstract, fuzzy Extracted system data: File size, Date modified etc Extracted from system Extracted measurement value data: E.g. Weight (kg), Blood pressure Extracted from text Specific identifiers data: E.g. email, mobile number Extracted from text Named real-world entities e.g. People names, Organisations Types & Categories e.g. roles, disciplines Events & Actions e.g. birth date, appointment, conference Characteristics & attributes e.g. colour, size Entity relationships, connections e.g. disease symptoms, causes Topics / Subjects e.g. Paediatrics, Surgery Sentiment, opinion e.g. positive, negative Knowledge graphs can help with more complex searches & navigation
- 44. Creating a DCAT metadata catalogues (RDF base & W3C recommended)
- 45. Adding metadata for datasets (with RDF) Save/Upload pipeline System metadata Final_final.csv Metadata form using metadata standards in the background e.g. Dublin Core Description Processing & indexing pipeline Lineage, versioning Upload Annotations, categorisation etc Usage notation Form populated: Automatically Semi-automatically Manually Or a combination User rating Auto-populated by referencing the KG Let users determine Dataset usefulness Reusable (FAIR) Comments User can modify before uploading
- 46. Buy versus Build • Buy • Quick (initially) • Lock in? • Development? • New ideas, processes, sources? • Application centric • Build • Agile build • Data centric, easier application innovation • In control: data capture to end user consumption • Behaviour & usage measurement / policing • KGs used across • Integrate also Data Sensitivity & Access control • Flexible for new data interoperability changes e.g. convert to RDF data • AI (ML & NLP) readyKnowledge Graphs AI
- 47. Thank you