How do you create a taxonomy for products?

Question

I have the broader question, which is what is the definition of 'taxonomy'. What does it actually mean to create a taxonomy for something... Then I'd like to apply that to an example. Lets say the project is a redesign of an Ecommerce website selling books. You have been tasked with creating a taxonomy for products. What steps as UXer would you take need to take in order to achieve this?

Answer 1

The first thing to remember about taxonomies is that there is never one correct taxonomy per set of elements.

Imagine having the product set {paper, cotton fabric, old wooden lineal}. For a tailor, the categorization within their taxonomies is: {paper, old wooden lineal} -> pattern making supply, {cotton fabric} -> clothing raw material. For a paper manufacturer, the taxonomy is: {cotton fabric, old wooden lineal} -> raw materials for paper production, {paper} -> end product.

So, a taxonomy is not so much about the products, it is more about the users. Each user has his own "correct" taxonomy, or more properly, each combination of a user and task he has to achieve will have its own correct taxonomy (when you have a nail, everything starts to fall in the "looks like a hammer" and "damn, useless" categories). So, how does the world work if everybody has a different taxonomy? There are multiple factors which allow you to get you close to a good solution:

• the more similar the users and their needs, the more overlap will there be between their taxonomies
• a site is not limited to one taxonomy
• navigating a taxonomy is not the only way to discover products

As a user experience specialist, you will want to first create a composite taxonomy which gives you the best overlap of the individual correct taxonomies of your users. This can be done with the usual methods for discovering user needs, this time focused specifically on taxonomies. I can't enumerate them here, this is a topic which cannot be exhausted in a single book, much less a post here.

The second thing to consider is that often, there are several good categories for a single product, each following from its own taxonomy. You should simply use them all at once. On a fabrics website, the same fabric can be listed under "Fabrics by purpose -> Clothing fabrics -> Winter fabrics", "Fabrics by material -> Natural fabrics -> Wool mix fabrics", and "Fabrics by color -> Weaved pattern fabrics -> Herringbone pattern fabrics". The user can navigate to it starting from each root. Amazon does something similar, look at all categories under which a book is found.

The third thing: You will never be able to create a composite taxonomy covering all the individual taxonomies of all users who will ever visit your site. And users need lots of time (and fail often) to find a product categorized by a taxonomy which differs a lot from their own. So provide good search in addition to taxonomies.

Answer 2

Definition

The word Taxonomy originated from the work of Carl Linnæus, who created an hierarchy of organisms in the 18th century. The word Taxon means a group of organisms.

Since then, the word has been used to describe a multitude of classification schemes, mainly ones with strict hierarchy. While definitions do exist, the term is often abused.

However, within information architecture many will agree on the following definition:

A need-based classification scheme that organises controlled vocabulary into an hierarchical structure.

And to elaborate:

• Need-based classification scheme - We classify entities based on the user needs (the term needs is used here rather than subject or topic; interest could also work).
• Controlled vocabulary - The terms (labels) are carefully selected.
• Hierarchical structure - A structure involving parent-child relationship.

Or in simple terms, a UX taxonomy is:

A tree structure of terms users expect that reflects user needs.

The Challange

UX taxonomists face 4 key challenges:

• Items - What is being classified? Sometimes this is obvious, sometimes research and data collection will reveal this.
• Needs - What are the user needs? Interests? How do they approach the domain? For instance, do cloth shoppers of site X prioritise brand over item type?
• Terms - What terms are used by users? For example, what electrical engineers call a Potentiometer, audio engineers call a Pot or a Knob or even a Rotary Fader; so which one is more common? Is the term Knob ambiguous within the domain vocabulary?
• Groups - What are the categories and in which category an entity belongs. Do envelopes go under Office Supplies or Paper Products?

UX Activities

In the traditional UX design sense the process involves:

Research

In order to tackle the challenges above, a few activities can take place (these assume the items to classify has already been identified):

• Open Card Sorting - users are asked to fill cards with a term describing a domain item; users are then asked to organise the cards. This exercise will reveal both terminology and hierarchy.
• Close Card Sorting - users are asked to sort cards with terms already on them. Will only reveal hierarchy.
• Interviews - the terms used can be picked; directed questions can reveal users needs.
• User Observations - can reveal terms and needs.
• Competitor Analysis - how others structured their domain? What terms did they use?

Design

This is where you (and possibly other experts) sit and sweat blood evaluating different schemes and considering the terms.

Here, a few things come into consideration. For example:

• Breadth vs Depth - will the hierarchy be wide (many sibling categories) or deep (many levels)? Breadth can impose cognitive load (having to choose between 15 different categories), where depth can make interaction cumbersome (7 clicks to get to what you want). If in your control, this is also context-dependant: Small displays (mobiles) will call for depth, where large displays (desktop wide-screens) will call for breadth.
• Order - are the items/groups logically ordered. For instance, in the 4 challenges above, wouldn't Needs/Items/Groups/Terms make more sense?
• Poly-parenting - can an item appear in more than one category? Doing so can help users but involve some complexities (eg, breadcrumbs are problematic).
• The Single Child - Comes to describe a parent with only one child. This is a clear-cut don't: You group more than one item; if a group has only one item, either remove the group or add more items (by means of analogy - what's the point creating a sub-heading in a book if it has no siblings?). Many people new to structuring will create such relationship.

Testing

Ask users to find an item within the hierarchy.

It is important to remember that a problem-free solution is next to impossible. More on this in the Yet Another Design Problem section below.

The Process

The following task model shows key actions, products and evaluation points in the process:

• First, we spell out all the items/objects need organising. We sometimes call these 'leaves' - hierarchy nodes that have no children.
• Then, we start grouping items into groups (categories).
• Then group the groups into higher-level groups.
• Ideally, the result of this is more than one hierarchy.
• We then evaluate, test and refine the various options, until picking up the best one.

To give an example, say we wish to create a taxonomy of the UX field. An extremely partial items inventory (but one that would suffice for this example) would be:

Next, we will start grouping items:

Notice that the following breakdown is equally possible:

You can already learn quite a lot from the last hierarchy. For instance:

• Observing users is under research, but it will also be part of a group called testing.
• Semantic Memory and Short Term Memory are not real siblings - the former should really be grouped with Episodic Memory, while the latter with Long Term Memory. So further grouping needs to take place.
• Shall we call it Short Term Memory or Working Memory?
• Is Human in Human Memory needed? Wouldn't Memory suffice?
• Is prototyping really just design, or can it also be part of (partial) implementation?

The next step will be to group the groups, so you may end up with something like this:

We can now test the taxonomy, how easy would it be to fit in:

• Analytics
• Creating Personas
• Episodic Memory
• Visual Memory (will require another sub-group under memory)
• Creating Task Matrix

Yet another Design Problem

One problem with creating a taxonomy is that it nearly always represent a complex problem; as such, there is no best solution - only best compromise.

As with design problems, the diverge-converge strategy often yields best results. In other words, it is important to derive at different solutions rather than focus on one.

In practical terms, the process should not commit to macro categories too soon - doing so can yield a flawed taxonomy. As different attempts are made, a specific scheme could turn to be landing itself to the problem better than others.

An Alternative (future) Approach

The problem with creating hierarchies is that the entities we classify pretty much always have more than one property/variable, thus they can belong to more than one group.

Both ontologies and topic-maps come to solve this issue. But these are more complex and aren't navigation friendly. However, one can take an hybrid approach when creating a taxonomy. The idea is to spell the properties (facets/classes/tags) of each item, and from the created property list derived at possible categories; like so:

While this is clearly more involved, it can be used to create both hierarchical taxonomies and topic-maps/ontologies. With the latter becoming more widespread (and possibly a requirement for SEO), it may pay to take this path.

Answer 3

To simplify, taxonomy is categorization. To create a taxonomy for something means to find commonalities that a user of that system would acknowledge are linked and categorize them accordingly.

Like anything UX related, you start with the user. It seems like your question more specifically asks How does one go about categorizing books for an e-commerce site? Well, "User-centric taxonomies" are simply categorizations made based on things like personas, web analytics, user testing and card sorting, to name the obvious big hitters. You'll have to follow these traditional avenues to accumulate the information necessary. After this information is gathered, you'll be armed with the info to make taxonomies that resemble an Amazon crumb like this:

There are of course considerations such as future proofing by creating labeling guidelines or limiting taxonomies especially for user created content sites - think Pinterest - so that new content isn't submitted without being categorized (you want to be able to find it, right?).

To summarize, you're really after research data because without a user-based framework, it's a shot in the dark. Just an example of how two different e-commerce book stores will categorize based personas/analytics/inventory (Amazon v. Oyster - subscription model):

Personas/analytics likely lead to different Editor's Picks from Amazon and Oyster's Picks/ Popular on Oyster. Oyster has a significantly smaller inventory so it naturally has broader categorizations with no filtering. For example, there is just a Business Essentials category in Oyster, but look at Amazon's business section and the amount of filtering possible.

Answer 4

What you're talking about is Information Architecture, the art of creating meaningful and worthwhile structures for content. I've worked on a few Taxonomies and there are techniques I have used that work well and some which don't. Card sorting, for example, I've tried to use several times but have found it unreliable.

The steps I tend to use are:

• Gather the items you are catagorising and any existing hierachy. I tend to do this in a spreadsheet with top levels as illustrations for communication.
• Indentify a list of key items across the spectrum of items sold. This should include best sellers through to hard to classify niche items. You can't test every item but need twenty or more items to test out any taxonomy you create.
• Gather site usage data you might have including real world experience of merchandisers if you have it.
• Do some user research / testing of an existing site and competitors. Testing competitor sites is a great way to find out things and is legal! You don't need to tell them. I've tried card sorting many times but yet to find it a reliable method.
• Build a hypothesis hierarchy (again I do this as a spread sheet) based upon what you know. Pass this by merchandisers if you have them.
• Test the hypothesis. Tools like 'Treejacker' allow you to test just the taxonomy! 10 users will be enough.
• Tune the hypothesis and retest.
• Put it live and, if you can, multi-variant test any parts of your tree you're unsure about.

Other tips.

• An item can appear in more than one place. Multi-merchandising is fine.
• Facets are vital for large shops and replace the taxonomy in many cases.
• Beware tagging, tags are not the same as taxonomy labels and facets.
• Most good shopping taxonomies are broad and shallow.
• BUT if something takes seven clicks to get to it's not a disaster if the user takes that route quickly! For example Wine often has far more facets than most items. Some books might need more facets as well.
• Gifts and promotions will need to be added into the structure somewhere in most cases.

Lastly get hold of:

Information Architecture for the World Wide Web: Designing Large-Scale Web Sites

Answer 5

Taxonomy as defined here http://en.wikipedia.org/wiki/Taxonomy is a study of classification of things or concepts. Taxonomy of products will be categorization and definition product hierarchy.

Taxonomy of a corporate structure, geographical locations, etc are defined similarly.

You can define them in the form a hierarchy tree, with their labels describing the semantics and their levels (in the tree) defining how they are grouped.

This link should give an idea (Assuming that hierarchy definition isn't done on a mobile or touch-based device)

http://support.f5.com/kb/en-us/products/wa/manuals/product/wa9_4policy/WA_Policy_9_4-05-1.html

Answer 6

It depends heavily on who is using it.

Many of our customers use taxnomies for their websites, which simply mirror the website structure to a tree.

This is, by all means, no viable solution.

The approach to this were filters.

We start with a generic solution like product categories or sitemaps etc.

Then we let the users browse for what they need and let them filter everything else away they don't need and save those filters. Sometimes this good enough.

After they did this for a sufficient amount of time, we check their filters and try to extract new informations out of them. Checking if their filters group items together from different branches and why.

This approach usually leads to more distinct orders.

Answer 7

the project is a redesign of an Ecommerce website selling books. You have been tasked with creating a taxonomy for products. What steps as UXer would you take need to take in order to achieve this?

https://vimeo.com/222586117 explains the use of taxonomies and how to merge many into one.

If you wanted to map the old taxonomy to the new one, and you had already designed the new one, you could import both into treemerge.io and create the mapping visually. Then you could run a script to assign all the book items to the new category with the mapping.