Sunday, April 23, 2017

Taxonomy Term Specificity

One of the challenges in creating or editing taxonomies is determining how specific the terms should be. This is a key issue in making a taxonomy customized for a certain implementation, which involves a unique set of content to be tagged/indexed and a certain set of users. Highly specific terms tend to be the consequence of deeper hierarchies. So, the decision of how specific the terms should be is also related to the decision of how many hierarchical levels of depth the taxonomy should be. Taxonomies that are organized into multiple facets, on the other hand, tend to have more limited hierarchy, if any, and terms that are not so specific.

Having taxonomy terms that are more specific than necessary inevitably means that there are more taxonomy terms than necessary. The larger taxonomy is more difficult to maintain both in currency and consistency. Terms that are more specific than necessary are also likely to be more specific than expected by the users and might get overlooked and not even used. If the taxonomy is searched, the users will not likely search for such highly specific terms. If the taxonomy is browsed, the users might stop at a higher-level broader term and be satisfied with that. Furthermore, users like to retrieve multiple results (content items or references) for a single search term, so that they can browse the list and evaluate what they want. Highly specific terms will match fewer content items, so retrieved results could comprise only one or two items per taxonomy term, which may not satisfy most users. Having a greater number of more specific terms can also lead to more inconsistency in the indexing/tagging, whether manual or automated. 

Having taxonomy terms that are not specific enough means that each taxonomy term is indexed to a relatively large number of content items, and the users may have to scroll through multiple screens of returned results and look at multiple items to find what they really want. The availability of additional filters or facets can help limit the results, though. Having terms that are not specific enough also makes it more difficult for users to “discover” potential related topics of interest, whether the terms have “related-term”/”see also” relationships between them or whether “related” terms are suggested by shared tagged occurrence among content items.

Taxonomists sometimes refer to term specificity as “granularity” or a taxonomy being “granular.” There is the irony that, although the scope and meaning of specific terms is granular/narrow/small, the terms themselves are not small. The “granular” terms tend to be longer, more complex, multi-word terms. If combining multiple concepts into a single term, such terms might also be called "pre-coordinated" terms. Following are examples of specific, granular taxonomy terms from different specialized taxonomies:

  • Possessed object access systems (in an information technology taxonomy)
  • Fingerstick blood sugar testing (in a health care taxonomy)
  • Standard manufacturing overhead cost (in a business taxonomy)

The taxonomist typically creates specific/granular terms, based on the concepts of sample content to be tagged. There may be a document with the phrase in the title, an image with the phrase in its caption, a product with this description as its type, a department with the phrase in its name, etc. Obviously, source phrases would need to be edited to become well-formed taxonomy terms, but they may still be multi-word, complex terms. Creating a taxonomy from scratch usually involves a combination of a top-down and bottom-up approach in the development of terms and hierarchical relationships. The specific/granular terms are the result of the bottom-up component of taxonomy development.

Taxonomies available for license might be appropriate in their subject area and scope, but chances are that their terms get either too specific or not specific enough for different implementations. Thus, if you choose to license a taxonomy, make sure your license allows you to customize the taxonomy so that you can either delete terms that are too specific or add more terms, as narrower terms to existing terms, that are more specific to suit your content

Creating or deleting specific terms is also part of periodic taxonomy maintenance. If a term, which has no narrower terms, is heavily used in indexing, it might be time to “break it up” be creating a few more specific, narrower terms so that the large content set is indexed and retrieved with more specific terms for more manageable result numbers. If, over a period of time, a specific terms has been applied in indexing very few times, or not at all, it should probably be deleted. The deleted term can be changed to a variant/nonpreferred term/alternative label for an existing broader concept. The specificity of a taxonomy should match the specificity of the content being tagged with it, and this can change over time.

Friday, March 17, 2017

Taxonomies as Knowledge Organization Systems

A taxonomy is a kind of controlled vocabulary. A taxonomy is also a kind of knowledge organization system. So, the question is: what’s the difference, if any, between a controlled vocabulary and a knowledge organization system? When I first heard of “knowledge organization system” I perceived it as merely a more academic term for controlled vocabulary. While it’s true that knowledge organization systems are discussed more in library and information science literature and courses than they are in corporate enterprises, there are additional nuanced differences between the two.

Controlled vocabularies comprise simple term lists, synonym rings (search thesauri), authority files, taxonomies, and thesauri. Knowledge organization systems comprise all of these, plus categorization schemes, classiļ¬cation schemes, dictionaries, gazetteers, glossaries, ontologies, semantic networks, subject heading schemes, and terminologies. As such, knowledge organization systems can be considered to be broader than controlled vocabularies, including all kinds of controlled vocabularies and more.

Yet, it’s not simply a matter of more types that distinguish knowledge organization systems. Knowledge organization systems include “schemes” that go beyond how the terms are organized and related to each other. Categorization schemes, classification schemes, semantic networks, ontologies present not only terms and relationships but also models of how information/knowledge can be managed and organized. These typically involve additional specifications and documentation on how they are to be used. There is indeed something to the name “knowledge organization system.” A “system” is more than just terms and their relationships.

As such, there is more discourse around knowledge organization systems than controlled vocabularies, per se (separate from discussions specifically about taxonomies or thesauri). Conference sessions of the Association for Information Science & Technology (ASIS&T) more often have “knowledge organization systems” in their titles than “controlled vocabularies.” There is even a professional association dedicated to knowledge organization systems, the International Society for Knowledge Organization (ISKO). There is no comparable organization for controlled vocabularies or just taxonomies or thesauri. ISKO holds conferences with sessions around the various issues of knowledge organization systems, including taxonomies. Recognizing that taxonomies are an important kind of knowledge organization system, the ISKO UK chapter co-sponsors the Taxonomy Boot Camp London conference

Taxonomies are not only included within knowledge organization systems, but they are also a part of the field of knowledge management. As a consultant, I worked with clients who managed taxonomies within their knowledge management services, headed by a manager or director of knowledge management. Also, at a consultancy where I previously worked, taxonomy consulting was part of the larger knowledge management consulting practice

I used to describe taxonomies as only a kind of controlled vocabulary, but now I will start referring to them as knowledge organization systems as well.

Sunday, February 19, 2017

Avoiding Mistakes in Taxonomy Hierarchical Relationships

Perhaps the most important issue in designing a hierarchical taxonomy is creating hierarchical relationships between terms correctly. This makes the taxonomy intuitively easy to understand and navigate by all kinds of users, regardless of whether they have had any training on using a taxonomy.

The basic principles of the hierarchical relationship are described in the ANSI/NISO Z39.19 and ISO 25964-1 standards for thesauri. As a quick summary, the relationship is created between terms in the following circumstances:

  • a broader term which is generic and a narrower term which is a more specific type of the generic broader term,
  • a broader term which is generic and the narrower term is a named instance (proper noun) of the generic broader term,
  • a broader term which is a whole entity and a narrower term which is an integral part.

It is the first, generic-specific type that is most common, but is also most prone to errors by those not experienced in creating taxonomies. Typical errors include confusing refinement and narrower terms, too closely reflecting the source content hierarchy, and creating narrower terms that are applications, uses, or examples of a broader term.

Confusing Refinements with Narrower Terms

We envision users browsing a hierarchical taxonomy from top down, from broad topic to more specific topic. A more specific topic is a narrower term (NT) of a broader topic. However, instead of providing more specific topics, the creator of a taxonomy might mistakenly provide refinements of the broader topic, which are aspects of the topic, but not actually narrower terms. A term that is an aspect or refinement is not a unique stand-alone term/concept, but rather it is meant to be used in combination with its parent term.

An example of such an erroneous hierarchy would be:

  Eye diseases

Diagnosis is an aspect or refinement of Eye diseases (and of other disease-type terms), and not a narrower term. A narrower term would be specific type of eye disease:

  Eye diseases
     NT: Glaucoma

A refinement term might not be as obvious as it is in the above example. If the same term, however, appears duplicated as a narrower term to different broader terms, but with a different implied/contextual meaning in each case, this should be red flag that the duplicated narrower term is really a refinement term. For example, the duplication of the term Waiver in a legal taxonomy as:

  Objections to evidence

  Right to jury trial

In this case, the duplicate narrower term should be changed to be specific in each case, such as: Objections to evidence waiver and Right to jury trial waiver.

Novice taxonomists might create such incorrect broader term-narrower term relationships because they have seen them formed as such elsewhere, such as Library of Congress Subject Headings plus Subdivisions or back-of-the-book index main entries plus subentries. A subheading or a subentry is not the same as a narrower term, because a subheading or a subentry only has usage and meaning in the context of the main heading it is associated with (appears under). A taxonomy narrower term, on the other hand, is not a different kind of term, but is rather a description of a relationship between terms. The meaning of a term in a taxonomy is constant and not dependent on its location in the taxonomy.

Too Closely Reflecting the Source Content Hierarchy

Some taxonomies are based heavily on certain text sources, such as the table of contents of one or a limited number of books or manuals, where the text is structured into units, chapters, main heading sections, subheading sections, etc. It is thus natural to make use of the structure of the text as a basis for the structure of the hierarchy. But there can be issues.

In the following example of a chapter and its headings from a textbook, greater hierarchical structure is needed for the corresponding taxonomy terms, and one of the topics (Units of Measure) does not belong within this hierarchy.

  Microbiology Laboratory
  --Microbiology Lab Personnel
  --Introduction to the Microscope
  --Units of Measure
  --Types of Microscopes
  --Laboratory Staining Methods
  --Culture Media

These concepts may appear in a taxonomy arranged hierarchically as follows:

  Medical laboratory technology
  NT: Laboratory equipment and supplies
       NT: Culture media
       NT: Microscopes
            NT: Microscope types
  NT: Laboratory personnel
  NT: Microscope use
       NT: Microscopy stains
  NT: Serology

  NT: Measurements and calculations
       NT: Units of measure

Another issue is that, even when the the hierarchy from the source is acceptable, the subheading-based terms are short, generic, and without context. An example is as follows:

  Eye Medications
  --Anti-inflammatory Agents
  --Antiglaucoma Agents
  --Local Anesthetics

The only correct narrower term above is Antiglaucoma Agents, as the other terms are not specific to eye medications. They could be linked as related terms instead.

Applications, Uses, or Example-Type Terms

Relying too much on certain text sources for the taxonomy may also result in erroneously creating narrower terms for the applications, uses, or examples of the broader term concept, because the text presents content that way.

Following are several examples:

  Web Applications
  --Tourism and Travel
  --Higher Education
  --Financial Institutions
  --Software Distribution
  --Health Care

  Decision making issues
  --Ethical conflicts
  --Information sources
  --Intraorganizational conflicts
  --Social influences

  Globalization challenges
  --Cultural differences
  --Economic risk
  --Political risk
  --Managerial limitations

Each of these so-called narrower terms are merely examples within the context of the broader term. All "narrower terms" could have other uses beyond the context of the broader term. To make the hierarchy correct, either:
1) the relationship should be changed from narrower-term (NT) to related-term (RT). This would be the case, if these terms can logically exist elsewhere in the taxonomy. Also, indexing of the concepts may require a pair of terms (such as Globalization challenges AND Economic risk),
2) the narrower terms should be modified and clarified, such as Cultural challenges to globalization, Economic risk challenges to globalization, Political challenges to globalization, and Managerial challenges to globalization. This would be the case, if these terms did not exist elsewhere in the taxonomy.

In conclusion, hierarchical relationships need to be constructed independent of any sources for terms, and they need to be universal and not subject to certain contexts.