Friday, November 24, 2017

Auto-categorization and Taxonomies



Taxonomies and thesauri are only truly useful if their terms are appropriately indexed or tagged to content. My path to taxonomist had been as an indexer, so I always value the importance of human indexers. Nevertheless, I must acknowledge that automated indexing, also called auto-categorization, is becoming increasingly common and important.

At the most recent Taxonomy Boot Camp conference (November 6-7, in Washington, DC), a trend I discerned was the increasingly commonplace use of auto-categorization (or at least machine-aided indexing) with taxonomies. Conference presentations didn’t state auto-categorization as something new but rather sometime more matter of-the-fact, and by the way, the software vendor used in this case is so-and-so. There were also sessions on artificial intelligence and taxonomy and on leveraging taxonomy management with machine learning. There is also a lot of interest in text analytics, a field broader than auto-categorization, which justified the first Text Analytics Forum conference co-located with and immediately following Taxonomy Boot Camp (which I, unfortunately, did not have time for).

When conference speakers and others state that automated indexing has been proven repeatedly in test comparisons to be more “reliable” and more “consistent” than human/manual indexing, while true, that does not mean it is better. Human indexing is certainly not as consistent, as two trained indexers will not index exactly the same way, but the way they differ is rarely so substantial. One indexer may add an additional index term. Another indexer may index with a slightly different, but related, term. Automated indexing, on the other hand, while consistent, is not as correct. Depending on the method, it can be approximately 20% inaccurate, indexing with completely wrong terms or completely missing the most appropriate terms. That’s where “machine-aided indexing” comes in, where indexing is initially automated, but a human quickly reviews the suggested terms, adding or deleting terms as appropriate.

The primary reason for implementing automated indexing is not so much to achieve consistent indexing, but rather to achieve efficient indexing. This is because the amount of content to be indexed in many organizations is growing too fast to be kept up with by manual indexing. Publishers of external content for subscribers have also transitioned to partial automated indexes or machine-aided indexing.

While enterprise search engines do not utilize taxonomies by default (but can be configured to make use of them), auto-categorization software generally uses some form of taxonomies. Search engines can function out-of-the-box without any taxonomies or controlled vocabularies, although a search thesaurus (a.k.a synonym ring) can significantly improve search precision and recall. Auto-categorization software, on the other hand, relies on “categories,” which can be simple controlled vocabularies or hierarchical or faceted taxonomies. Thus, as auto-categorization is gaining wider adoption, the need for taxonomies to support them is also growing.

Automated indexing technologies have not advanced significantly in recent years, but there have been improvements in auto-categorization software by effectively combining more than one technology method within the same software product. The main technology methods are (1) rules-based and (2) machine-learning. Regardless of the method, automated indexing is still not fully automated. Humans are required to put in time and effort beforehand to either write or edit rules for each taxonomy term, or to provide and test training sets of sample documents to index for machine learning. These could be dedicated roles or additional tasks to be performed by the taxonomist.

Auto-categorization is also becoming more common, because software products that effectively combine taxonomy management with auto-categorization have become more established and better integrated. Although there are many organizations which continue to use distinctly separate software for each of taxonomy management and auto-categorization, organizations newer to taxonomy adoption prefer to have a single solution. Synaptica is the one major taxonomy management vendor which does not yet include fully integrated auto-categorization, and they are very actively working on incorporating the technology. I have separate chapters in my book, The Accidental Taxonomist for software for taxonomy management and software for auto-categorization, but in my second edition I ended up repeating more vendors in both sections.

Saturday, October 21, 2017

Taxonomies for Specific Business Needs

Designing controlled vocabularies to meet specific business needs was the topic of my latest conference presentation at Taxonomy Boot Camp London on October 17. There are two aspects to this topic: (1) the type of controlled vocabulary to choose, and (2) whether to have the same controlled vocabulary or distinct controlled vocabularies to serve different business needs.

For choosing the type of controlled vocabulary, the most common choices are a thesaurus, a hierarchical taxonomy, or a faceted taxonomy. It is also possible to have some kind of combination or hybrid type of these. I’ve discussed the difference between taxonomies and thesauri in previous blog posts, “Taxonomies vs. Thesauri”  and “Taxonomies vs. Thesauri: Practical Implementations.”
So, now I will focus on whether to have the same controlled vocabulary or distinct controlled vocabularies to serve different business needs.

What are different business needs? Taxonomies may be needed to make different kinds of information organized and easily searched or discovered and retrieved by different users, including:
  • Internal documentation, including policies and procedures, market and product research, etc.
  • Digital assets for content managers to reuse in publishing content 
  • Product information, such as a product catalog for ecommerce, for customer
  • Curated, premium content for subscriber
  • Informational content for the public
 While different organizations have their own needs, the same organization could have more than one business need for a taxonomy, such as an internal use and an external customer-facing use. An organization in the business of publishing content, may even have quite different published products for different users and purposes and consider each of those as separate business needs.

Taxonomies are versatile, so it is possible and worth considering having a single, master taxonomy serve all business needs, with terms classified for different uses. Terms managed in a taxonomy management system can be tagged with a category assignment as to which use they are for, such as some for an internal use, some for an external, and perhaps many of them for both.  You determine the type of category, let’s say “audience”, determine what audience types there are, such as “internal” and “external,” and set that up in the categories option of your taxonomy management software. Then you assign the categories, as appropriate to each term. This method works if the same terms and the same structure are being used in both cases, with one use having more specific terms in some areas. The other use may have less specific terms, or also more specific terms in other areas.

The method of using the same taxonomy for different uses, designating use by categories on terms requires
  • that when a concept in the taxonomy has more than one use, that the same preferred term label is used for the concept in both/all cases
  • that concepts/terms in the taxonomy have the same relationships to each other
Sometimes, however, different business needs require different preferred labels for a concept, such as “Customers” vs. “Clients.” It is possible to maintain multiple preferred labels for a concept, if you manage them as you would manage multiple language versions of a term in a multilingual taxonomy, but this is more complexity than necessary when only some of the terms have different preferred label.

If you want to maintain links between equivalent terms, whether they have the same preferred labels or not, in different business-use versions, it’s not necessary to maintain them in the same taxonomy akin to multilingual versions. Rather, if you created two separate taxonomies, you could still set up inter-taxonomy links between the equivalent terms. This is not necessary, but might be desirable.

Whether to maintain one or more taxonomies also depends on the size of each. If one of the business use cases requires only a small taxonomy, of a couple hundred terms or less, then it is not too much trouble to maintain distinct taxonomies for each.

Saturday, September 30, 2017

Vocabulary Management Issues



Issues in Vocabulary Management” is the latest Technical Report (TR-06-2017) published by the National InformationStandards Organization (NISO), approved on September 25, 2017. I had the honor of serving on its working group, specifically on its subgroup for Vocabulary Use/Reuse.

The most significant NISO publication for controlled vocabularies is ANSI/NISO Z39.19-2005 (R2010) Guidelines for the Construction, Format, and Management of Monolingual Controlled Vocabularies, which is referenced several times in TR-06. ANSI/NISO Z39.19 focuses on how to design and create controlled vocabularies (especially thesauri and taxonomies), whereas TR -06 addresses issues in the use of controlled vocabularies. Furthermore, as a Technical Report, rather than a Standard, this 49-page document does not contain requirements, but rather serves an informative purpose. It does have a page of recommendations, though, which are for a vocabulary’s definition and attribute types, its best practices for documentation, and its licensing or provisions for use and reuse.

Over time, the need to create new controlled vocabularies from scratch diminishes, as more vocabularies come into existence, especially those that are made available for sharing or licensing (see my blog post Directories and Databases of Published Controlled Vocabularies) but the need to maintain, revise, and reuse them grows, so this Technical Report serves a valuable role.

What are the “issues” in vocabulary management? They could vary, based on the organization and implementation, but this document considers three areas of

  • Vocabulary use and reuse, dealing with permissions, licenses, maintenance, versioning, extending and mapping vocabularies.
  • Vocabulary documentation, dealing with governance issues and how to document vocabulary properties.
  • Vocabulary preservation, dealing with issues of abandoned or “orphaned” vocabularies, which is especially the case of vocabularies developed by nonprofit organizations which have lost their funding to maintain them.

These issues are relevant to both proprietary controlled vocabularies, which may be reused through licensing agreements, and publicly available vocabularies, which are shared and reused increasingly through linked data on the web, or more specifically the Semantic Web and the Linked Open Data environment.  For publicly available or open vocabularies there are also the issues of simply finding or discovering suitable and sustainable vocabularies and evaluating them and then the communication between the vocabulary owner and user.

TR-06 takes a somewhat broader view of “vocabularies,” not just “controlled vocabularies,” but also including ontologies, unstructured term lists, terminologies, synonym rings, etc. I explored these differences and definitions in detail in my blog post Vocabularies and Controlled Vocabularies, which I wrote shortly after starting work on the NISO working group. The vocabularies of concern of TR-06 also include element sets, which comprise metadata properties/fields and not merely the controlled vocabulary terms/values within those properties.

TR-06 does not seem so much as a “technical report.” It also includes several real-life examples and use cases. To a certain extent, it explains by example.  Appendices include a glossary of terms with extensive definitions; a descriptive list of vocabulary directories, repositories or collections (something that I worked on); a list of free and open vocabulary tools (far more extensive than those I described in a previous blog post Free Taxonomy Management Software); and a list of additional resources with links, besides its bibliography, making this quite a valuable resource.

TR-06 “Issues in Vocabulary Management” will now be added to my list of recommended resources for controlled vocabulary and taxonomy management, and I hope that many of those who manage taxonomies will take a look at it.

Tuesday, August 29, 2017

Taxonomies in SharePoint



Controlled vocabulary metadata, including hierarchical taxonomies, has been supported in SharePoint since its 2010 version, and its use and features have been enhanced is succeeding versions of SharePoint. While it’s not technically difficult for users to create taxonomies and apply their terms to content items in SharePoint, developing a metadata/taxonomy design and application strategy is definitely a challenge.

The distinction and overlap between metadata and taxonomies was the topic of my previous blog post, "Metadata and Taxonomies," and it is very relevant to SharePoint. Controlled vocabularies or taxonomies used to tag content are referred to in SharePoint as “managed metadata.” This designation is indeed accurate and fitting. Some, but not all, metadata is in taxonomy form (hierarchical structures), and in SharePoint it is managed/controlled in a central way, where permissions on who can change or add to the metadata terms may be limited to a smaller set of users than those who may tag content with the metadata. “Managed Metadata” is something you will hear about in documentation, but in the SharePoint application itself, what you want to work with is “Term store management,” grouped with other Site Administration settings under “Site Settings” (under the gear symbol in Office 365 SharePoint). Terms are grouped into “Term Sets” (top-term hierarchies or facets).
Questions to consider in taxonomy design in SharePoint include:

  • To what extent will document libraries (virtual folders) be used to categorize content within a site, and would proposed subfolder names be better suited as metadata terms for tagging?
  • Will the primary use be for filtering lists of documents in place, within an open document library, based on metadata selected for the various “columns,” or will the primary use be for refining search results, based on metadata selected in the left-hand margin refinement panel after executing a search?
  • How many Term Sets should be created and how many and which metadata fields in total should display to the users, either in columns or as search refinements.
  • When should a Term Set be a flat list and when should it be created as a hierarchy, and how deep should the hierarchy be?

Use of document libraries vs. metadata tagging 

 

SharePoint supports the creation of a hierarchy of nested folders within libraries within sites. So, it may be tempting to start of creating such a “taxonomy” of categories for content, especially if migrating content over from a shared drive where such folders had been used. However, tagged metadata has many advantages over categories of folders for finding and retrieving content.
A content item may be tagged with more than one term from the same Term Set if it deals with more than one topic or if it falls into more than one category type, whereas putting a document in more than one folder can lead to version control issues. (It’s true that you can put a document in one folder and a link to it from within another folder, but this is not easily remembered to do, nor does it look as “clean.”)

You can create Term Sets (as facets) each for multiple ways to categorize, such as by document type, function, audience, topic, etc., serving as facets, and then tag a content item with terms from each, whereas folders don’t deal well with mixed methods of categorization, and you are forced to choose one method of categorization.

  • Tagged metadata allows you to filter a large set of content in place to quickly narrow results to what you want, whereas folders require clicking down through multiple paths, taking more time to find desired content, which is in different places.
  • Tagged metadata can also be implemented as search refinement filters, also known as faceted search.
  • Tagged metadata terms can have synonyms, helping users find what they want by different names, whereas folder names cannot have synonyms. 

Thus, what had been labels for folders on a shared drive should most likely be changed to terms in taxonomy Term Set. Whether you should have any document libraries, or just a few without subfolders, depends on the preferences of your users, but I don’t recommend the creation of subfolders. 

Filtering on columns vs. refining searches

 

The same Term Sets may be used for both column filters and search refinements. But typically, the implementation of managed metadata in SharePoint is either primarily for one or the other purpose, and the other use may not even be set up. Generally, if the managed metadata is going to be applied to documents within a single library or one site, on the order of tens or hundreds, then column filters are desired; if the managed metadata is going to be applied across multiple sites on thousands of documents, then search refinements would be used.

If unsure whether to promote filtering on columns or refinements on search, consider that filtering columns will always get more accurate results, but metadata has to be consistently applied.  Out-of-the-box search in SharePoint will retrieve documents with the word or phrase anywhere in the document. The idea behind this is to get search set that is larger than needed and not miss anything, and then the user can refine the search result with the various refinements. So, the results of search are not as accurate, but there will be results, even if metadata tagging is incomplete.

Knowing how the Term Sets will be used can have an impact on the wording of terms and the extent of use of hierarchy. Both columns and refiners have limited width for term names to display. The user can easily adjust column widths to accommodate long names, but the refinement panel width cannot be widened by the user. The use of columns also makes it desirable to keep terms to a limited length within a given Term Set.  Refiners indicate hierarchies of terms by default to the user who is searching content, whereas columns do not indicate any hierarchy in the default view.

Number of terms sets and metadata fields

 

There is no point in creating a Term Set if it’s not going to be displayed to the users for filtering or refining, and too many metadata fields take up horizontal or vertical space, are a burden to tag, and make the user experience of searching or filtering too complicated. So, you need to consider what would be truly useful, and not merely possibly nice to have. Just two Term Sets, such as Document Type and Topic, may be sufficient. In addition to the managed metadata that you create, there will be other metadata fields desired for filtering or refining, such as date, author, and format type, and perhaps uncontrolled keyword tags applied by users. In the case of columns, there will always be the document title taking up a column and considerable horizontal space as well. 

The default columns in SharePoint are “Type” (file format) “Name” (filename), “Modified” (the date the file was uploaded or the last time any of its properties were updated, not the date the file itself was modified) and “Modified By” (the person who uploaded the file or last updated its properties, but not necessarily who actually modified the file). The default search refinements are the same, excluding title: “Result type” (file format), “Author” (an even worse misnomer for Modified by”), and “Modified date” (often displayed in a graph form). If you believe such information is not that valuable, you can remove these columns/refinements, especially when you plan to add other columns/refinements, which will take up horizontal or vertical space. 

I would recommend no more than 4-7 total metadata fields, including those that are not based on managed metadata. You should avoid having more metadata as columns, along with the document titles, than can fully display horizontally, so as not to require horizontal scrolling. Search refinements, on the other hand, by default display sample high-use terms under each refiner, so typically no more than three refiners display in the left margin without vertical scrolling. Vertical scrolling is expected and acceptable to a limited degree.


Term Sets as flat lists or hierarchical taxonomies

 

SharePoint makes it easy to create hierarchies within Term Sets by simply right-clicking on a selected term and selecting “Create Term” from the context menu. Some people might thing that since the Term Store is for taxonomies, and taxonomies are hierarchical, hierarchies should be created if applicable. However, hierarchies are only helpful for navigating the taxonomy if the taxonomy is sufficiently large. If you set up multiple Term Sets, each used as a facet in combination with others, they each don’t need to be very large. Furthermore, the types of content most people store in SharePoint tends not to need extensively large and detailed taxonomies as might be needed in a content management system or digital asset management system

My rule of thumb is up to 12 terms should be on the same level before considering creating any hierarchy, but it could go up to 20 or so, and even more if the list of named entities/proper nouns.  Also, if you do have hierarchies, consider keeping them relatively shallow, such as to only two levels, instead of three. Even if a hierarchy is technically correct, it does not mean you have to set it up that way.

If you need only a short flat list of terms, you might consider not using the Term Store at all, but rather create the list as "Choice" type of column. This is easier to implement, but the terms would limited in their use to filtering and sorting the column, and could not also be applied to search and navigation.