Vocabulary Interoperability using Linked Data : Principles, Capabilities and Some Examples. ALA LLD IG 20140126

Editor's Notes

• #3: This concept is at the heart of my talk. (As I thought about this idea it occurred to me that it is not revolutionary at all – especially to a librarian. I have a habit of relating every new field I become involved in back to traditional librarianship. Very few advances in information management are truly revolutionary to a librarian. Take “vocabulary interoperability” for example. Consider a traditional “library”. Isn’t it about “book interoperability”? Isn’t it about the ability for books to be adapted (selected, purchased, circulated), transformed (read, used to create new knowledge), and interlinked (arranged and made accessible according to common subjects and classifications), while--and this is key--never actually changing the books themselves? )
• #4: The purpose of this slide is to position the Semantic Web and Linked Data against an imperative of living in networked world. Too often Linked Data appears as a solution looking for a problem. No—the problem, the opportunity, is here. Linked Data is a solution designed to address the issues facing every information professional, including librarians and taxonomists. Organizations have been evolving from being very self-contained to becoming, what Gartner calls, the “hyperconnected enterprise”—a highly fluid, highly connected collection of organizations, where coordination is key—not control. What does a hyperconnected enterprise use to manage this network? Information of course. How do controlled vocabularies fit into this picture? 1. One Enterprise: Simple if you’re one organisation, with good control of your information. Create an enterprise taxonomy, or set of enterprise vocabularies, and have everyone use it; establish a governance structure for maintaining it. As I mentioned before I represent R&D on the AstraZeneca “enterprise taxonomy” – it’s a valuable resource used in our enterprise systems like SharePoint, Search, and digital asset management systems. We still need to manage our own information well. 2. Connected Enterprise: Of course, you need to use external information sources, which use their own custom vocabularies And you’ll need to deliver information to regulatory agencies that require the use of certain vocabularies (e.g. MedDRA, ICD-9/10, and other standards) Maybe another company has an enterprise taxonomy or product thesaurus – we can integrate those; give us a year. One of my first big projects at AZ was intergrating three product literature databases – each with their own thesaurus. We, of course, built an entirely new one. 3. Hyperconnected enterprise: But these other partners, universities, companies, divisions, subsidiaries… spread across the globe….partners come can partners go Small biotech – “Enterprise vocabulary? We’re too small – we don’t just use terms that we need; it’s not complicated.” Big pharma partner or regulatory agency – “Over here, yes we follow standard vocabularies…they’re not the same as your standards, but… “ Information supplier, database provider -- “Our proprietary databases use specially designed vocabs optimized for our software – you should really be using our software, you know. We have all the information need”. 4, 5. And this is not going to stop This is why can’t keep recreating vocabularies, or going through massive mapping efforts. It’s unsustainable. It’s the same driver that led to the world wide web – coordination trumps control. We need a way to for our vocabularies to be “interoperable”, where they can be linked to each other, while still retaining their native integrity.
• #5: How will we accomplish “vocabulary interoperability”? We need some basic capabilities. These were brainstormed at my organization. We don’t create a new vocabulary where one already exists.  Our preference is for authoritative, trustworthy, well-managed sources.   Standard criteria used to evaluate resources – authority, accuracy, currency, point of view (objectivity), coverage, relevance, format. http://library.uwb.edu/guides/eval.html , http://www.noblenet.org/merrimack/FYS_eval.pdf We can extend those sources if we have to, but even then we want to extend them by making associations with other authoritative sources.  Whenever possible we will leverage cross-vocabulary mappings already available publically.  When necessary, we will extend or map an external vocabulary with internal or bespoke terminology. This could include mapping obvious synonyms between vocabularies, or could entail more nuanced relationships. For example – mapping a published disease vocabulary to proprietary database value set, or list of values from internal database. We will not corrupt authoritative vocabulary sources.  It will always be possible to identify the source vocabulary, including what version, in the network. We don’t take a page out of this book and paste it into another, cross out a section here, add one there to the point where we can’t put the books back together again – we don’t do that with vocabularies either. But we do repurpose these vocabularies – so in a way we might need to pull them apart. Perhaps a deeply hierarchical taxonomy just needs to be rendered as flat list. But the semantic web promises us to enable this without corrupting the source. We will obtain vocabularies directly from the authority that produces them whenever practical. “directly” is the key word here. We’ll talk about this later. Vocabularies will be managed in RDF.  Preference is for this “SKOS” standard I’ve mentioned before. SKOS is a lightweight standard that can be applied to any controlled vocabulary.  Focusing on a simple standard allows us to standardize our curation, management, and delivery processes.  [In many cases, this requires doing our own SKOS conversion.] [aside: in some ways the nature of “published information” is becoming more like the early days of copying, commenting, annotating, all at the same time.]  
• #6: In addition to principles, we need capabilities. Again, these were brainstormed at my organization: Vocabulary Mapping. Reusing, creating, and maintaining the relationships between vocabularies holds the power of the “vocabulary network”.  We need to develop the approach, processes and tech for light-weight, sustainable, semantics-based vocabulary mapping.     How to map two or more vocabularies together using Semantic tech Facilitating the reviewing of automatic mapping - approve, reject, adjust, amend mapping Quality measures (e.g. single synonyms can’t be mapped to more than one parent). Leverage mapping already done in published vocabs. Publish mapped “bridging” vocabularies to customers. Vocabulary modeling and conversions. The vocabularies in the network need to be accessible as “Linked Data”.  And, until external suppliers deliver or make their vocabs available as linked data, we will need to be extremely proficient in converting them from their native formats. Vocabulary Slicing. Sometimes a vocabulary consumer – system or project – needs only a subset of a vocabulary that is relevant to their needs. For example, only the diseases that are of interest to a particular project. Rather than create a duplicate – we need to identify terms or branches of relevance, and make those available--all while retaining their link back to the authority, so that updates to the authority can continue to be leveraged (new synonyms, new mappings, new child terms, new notes). Vocabulary Version Management. As our focus is the reuse of already published vocabularies, so successfully managing the impact of new versions on the network is critical. Comparing new version to existing version Identifying impact of vocab changes to mapped vocabs, consuming systems or related datasets. Once we identify the impact – how do we addressing impact.    Vocabulary Inventory. Managing a newtwork of vocabularies requires a sophisticated inventory. Ideally managed as linked-data itself, the inventory can directly inform queries using the vocabulary network. For example, tell me all the vocabularies that cover concept X, the datasets that reference those vocabularies, and how to access them. There are developing standards in the semantic web specifically for managing these types of inventories.
• #7: These examples assume a basic knowledge of Linked Data standards and principles, RDF, etc., and a basic knowledge of SKOS. See extra slides for some additional background on these standards. NCIT has a unique URI minted by the owing authority. Once we can uniquely identify the concept we can start to record some basic things about it: It’s a concept It has a preferred label It has synonyms – obviously many more than than this one. In has a definition. It has one or more parents – listing two here to show how a polyhierarchy would work.
• #8: Here’s the same concept in MeSH – D001749. Here we’re using an ID assigned in the NCBO’s BioPortal (http://bioportal.bioontology.org) Once we can uniquely identify the concept we can start to record some basic things about it: It’s a concept It has a preferred label It has synonyms (“Entry Terms” in MeSH) – obviously many more than this one. In has a definition. It has one or more parents. Listing two here. A couple critical points here: If you look at the full record for this concept in Mesh, it does in fact have the NCIT preferred label as a synonym. But NCIt does not have the Mesh preferred name as a synonym. Note especially that alternate labels of “Cancer of the Bladder” in MeSH and “Tumor of the Bladder” in NCIt are unique to each vocabulary. But – what if we know that mesh:D001749 is the same concept as ncit:C2901!
• #9: What if you’re searching two carefully indexed repositories – one indexed using MeSH and another using NCIt. You want to produce a joined up view of these two repositories, leveraging the power of the two separate indexing methods? What if you’re searching across multiple repositories and you want as rich a synonym set for “Bladder Cancer” as you can get. As we just saw – by using only one, you could end up missing things.
• #10: Here’s how we use Linked Data to do really interesting things. This is a key slide. New Triple: ncit:2901 skos:exactMatch mesh:D001749 By creating one additional “triple” that identifies the match between the two concept URI, we’re saying that the concept called “Bladder Cancer in NCIt is the same thing, the same concept, as the concept called “Urinary Bladder Neoplasms” in MeSH. We can then exploit that new relationship by creating, essentially, a rule: SPARQL rule: We haven’t talked about SPARQL – but it is akin to SQL, in that it is the language with which to query and manipulate Linked Data in the semantic web New Inferred Triples: This three line SPARQL command takes that new relationship and reasons that the preferred labels or synonyms in the matched MeSH concept can be inferred as additional synonyms (in red) in the NCIt concept. It would do this for all the concepts in the NCIt that have been matched to Mesh terms. Those terms “reasoning” and “inferred” are very important. First you’re making a rule that a computer can use to infer additional information. Second--that’s all this has to be--an inference. We are not necessarily hard-coding additional data. We are not manipulating the source vocabularies at all. Every inferred triple can be managed completely separately from either source vocabulary!
• #23: Systems Librarian Starting around 1989 was responsible for and helped implement several library-oriented systems. While working at OCLC I had my first detailed experience with a thesaurus management system. I helped write a module that integrated MARC authority records with the thesaurus manager. Information Architect At AZ, around 2006, I led a team of librarian/webmasters that managed all the webcontent for the global library organisation. Taxonomist While working with enterprise content, I became involved with the global AZ team responsible for the AstraZeneca Enterprise Taxonomy. R&D Vocabulary Capability lead I first took on this role of vocabulary capability lead within the “R&D Information” department, working on taking data interoperability to the next level. It was clear that leveraging semantic web technology was the most sustainable way forward for vocabulary management.
• #26: Adding structure to unstructured text; essentially tagging unstructured content with vocabulary term identifiers, by looking for strings that match a wide-array of synonyms, and by other more sophisticated algorithms. Adding additional structure to already indexed content so that we can organize information in ways specific to our requirements. Adding vocabulary metadata to the content or to a search index to connect information across repositories of information by connecting the concepts embedded in that information.
• #27: In thinking about how to develop and deliver this service based on these new approaches, we’ve begun drafting some role descriptions.
• #28: I’ve been using the terms “semantic web” and “linked data” interchangably. The term “semantic web” has fallen out of favor among it’s advocates who instead refer more often to “linked data” – which sounds much more practical and obtainable than the futuristic vision touted in the seminal 2001 Scientific American article by Tim Berners-Lee, Jim Hendler, and Ora Lassila. 1. A web of data – not pages. Pages need to be read by a person, or theoretically an AI agent mimicking a person, in order to obtain elements of data. For example, I want to learn the times that local bars offer “happy hour” – I need to find the web page for each bar, scan pages, follow links, and finally read content that says something like “we have happy hour M-F from 4-7”. It doesn’t matter that that information might originally be in a database that builds the page. Because to interpret the information I need to read the page. Linked Data challenges us to use the principles of the web--unique identification or resources i.e. URLs, and links between those resources--to the data itself. URIs, RDF, and Ontologies URI – “uniform resource identifier”. URL’s are “uniform resource locators”--a type of URI used to uniquely identify pages and locate them. You “point” your browser to them to locate the unique page. In the semantic web, elements of data are also given URIs--using the same syntax. For example The Artful Dodger might have a webpage with a URL. But in the semantic web The Artful Dodger itself--as a thing, not a page, would have a URI. RDF – the most basic statement about data--subject/predicate/object--a “triple”. [motion with hands] For example The Artful Dodger (subject), Is a (predicate), Bar (object); The Artful Dodger (subject), Has a (predicate), Happy Hour (predicate). Ontologies – collections of information about classes of things and their relationships. For example, in this context - a Bar is a type of Restaurant. If I know that the Artful Dodger (s), Is a (p), Bar (o), then I can infer (an important word) that the Artful Dodger (s) is a (p), restaurant (o). That’s all I’m going to say about the fundamentals – let’s quickly see how they apply to our vocabularies, and specifically to our core capability of mapping two vocabularies together.
• #29: Concepts are Things Just like The Artful Dodger is a thing. The concept of a “Bar” is also a thing. Concepts have URIs Things are identified on the web with URIs A vocabulary is about concepts, properties, and relationships Obviously. In fact, it’s interesting to think of vocabularies themselves as little webs of knowledge. Well, they need not be little webs of knowledge unto themselves. They can join the larger web of data. SKOS – “Simple Knowledge Organisation System”. SKOS is a standardized set of classes and properties for representing “knowledge organisation systems”, aka vocabularies. Unlike other thesaurus, etc. standards, the fact that SKOS is Linked Data, means it allows for the management of vocabularies in a distributed, linkable, interoperable, way. In other words, as part of the open-ended, web of data. Lets’ walk through this diagram with a biomedial example using SKOS … Bladder Cancer (S) is a preferred term of (P) Bladder Cancer (O) Bladder Cancer (S) has an alternate label – synonym of (P) Bladder Neoplasm (O) Bladder Cancer (S) has a definition of (P) this… (O) Bladder Cancer (S) has a broader term (P) of Cancer (O) (which has it’s own properties) But it’s when we start mapping concepts between vocabularies that we really get the power we need.
• #30: I want to just outline some very common SKOS terminology as it will help with the following examples. skos:Concept skos:ConceptScheme As mentioned every “term” in a vocabulary is a concept represented by a unique URI. A ConceptScheme is generally used for a particular thesaurus, or part of a thesaurus. Concepts belong to one or more ConceptSchemes. skos:prefLabel skos:altLabel skos:definition These are classic thesaurus properties – prefLabel: preferred name for the term, altLabel: entry term, “see from”, “used for”, synonym, etc. definition – self explanatory! skos:broader skos:narrower skos:related These are the essential pieces to build a hierarchical or ontological framework, within a single vocabulary. skos:exactMatch skos:closeMatch Everything above is within a single vocabulary. These, instead, are used to relate Concepts across vocabularies. e.g. Bladder Cancer in this vocabulary is an exactMatch to Bladder Neoplasm in this vocabulary.