Product Categorization (Taxonomy, Ontology)

By , October 27, 2010

Creating the right structure of categories and attributes for the products in your online store is a critical step that many online merchants ignore.

I was disappointed today by an online merchant whose web site made it impossible for me to identify which product to buy, because the web site didn’t organize products into meaningful sub-categories. The “search” function didn’t help, because the merchant didn’t include adequate information in product descriptions.

Improper categorization makes it difficult for consumers to find products and to confidently make purchase decisions. This results in lower conversion (purchase) rates, lower return-visit rates, lower repeat-purchase rates, and higher product-return rates.

Many e-commerce businesses would be improved by the addition of more pinpoint categories, intelligent application of overlapping categories and attributes, the selective addition of qualitative and quantitative scores for many attributes, and the addition of relationship information among products and attributes.

I’ve worked on taxonomy/ontology and categorization in several ecommerce projects in the past decade, including:

  • Categorizing and tagging movies and posters for a movie poster e-tailer;
  • Categorizing and tagging famous quotations for a startup print-on-demand poster merchant;
  • Categorizing and prioritizing lesson-plan resources for a directory web site, monetized using performance-based advertising (affiliate programs).

In each example, I was responsible for assigning multiple categories, attributes, and attribute-values to each of tens of thousands of distinct works and products.

In my past categorization work, I’ve sought to create more useful overlapping “pinpoint” categories and sub-categories, and in each project I’ve also assigned qualitative and quantitative rankings within each category. I’ve also helped merchants to license taxonomies and ontologies (and additional content) from third parties, and to integrate this third-party data with the existing ontology and  in-house content.

My work has included the initial design (and periodic revision) to the ontology/ taxonomy (structure & content of categories & sub-categories and relationships), as well as thousands of hours spent assigning multiple categories and tags to individual products.

With the right information, customers could more quickly zoom in on the products that they find appealing, and could then find related products. These satisfied customers are more likely to purchase, to re-visit and to re-purchase; they would also be more likely to subscribe to receive notifications of new products that match their preferences.

It’s important to recognize that there are “multiple dimensions” for categorizing products in most businesses, which means that the merchant must either design a rigid database solution that includes separate fields for every product attribute, or else choose a ontological system that allows a single product to be tagged with multiple attributes, values, and relationships.

Properly designed, your taxonomy/ontology might also allow you to implement some relatively advanced marketing strategies (such as dynamic display advertising) which would otherwise require considerable manual effort.



A note on terminology: The words taxonomy and ontology have different meanings; put simply, a “taxonomy” is a hierarchical structure; “ontology” is a broader concept in which “objects” are defined with various classes, attributes, and relationships. Taxonomy is hierarchical; ontology is not. The term “taxonomy” was first used to describe the biological classifications of organisms; the term “ontology” originated in philosophy (metaphysics).

[The right-margin of this “blog” page contains two classification systems: a mostly-hierarchical “category/topic” system (in the pull-down menu at the top of the right column) plus a non-hierarchical “tag” or “tag” system. On this site, tags and categories are merged into a single “tag cloud” (reflected in the “tag cloud” at right). While the category system is arguably a taxonomy, the tag system is  merely a “classification system,” which is neither a taxonomy nor an ontology.]

It takes considerable time and effort to create a meaningful, useful, and relevant taxonomy and/or ontology for a specific use. Wikipedia is (arguably) a taxonomy; Amazon.com is built on a collection of taxonomies and ontologies; Google Search is actually built on a collection of ontologies.

Many folks view product categorization as a “taxonomy” (and out of habit, that’s the term that I usually use), but when implementing a meaningful e-commerce system, a hierarchical taxonomy is far too limiting, and the best implementation is via an “ontology.”

6 Responses to “Product Categorization (Taxonomy, Ontology)”

  1. Ondrej Zamazal says:

    Thanks for this interesting post. I am very interesting in this stuff. Could you give me some hints where I could find taxonomies/ontologies for some eshops? I would like to try some kind of matching between them. Thanks, Ondrej

  2. Mark Welch says:

    Ondrej, it’s fairly easy to find relevant resources by just doing relevant searches on Google, like “ecommerce ontology” or “product ontology.” I don’t know your specific project or goals, but I must warn you that much of the writing about ontologies is very complex and abstract (frequently written by authors with limited English skills), and many articles and web sites contain more dead (404) links than valid ones. Your research will often lead you to web sites that require substantial payments to view a particular ontology or taxonomy. You’ll also be guided to download and install a variety of software tools, some promising easy access to pre-defined ontologies or taxonomies, only later showing the price tag. There’s a lot of disappointment out there.

  3. suo says:

    Mark, it is an informative article. From a computer science point of view, ontology is also hierarchical.

  4. Mark Welch says:

    Suo, I disagree. My understanding is that an ontology need not be hierarchical. I do see that Wikipedia’s entry for “ontology” does identify an ontology as hierarchical, and its entry for “hierarchy” includes “overlapping hierarchy” in which an element can have multiple parents. However, I don’t think a traditional hierarchy allows for recursion or looping (in which one element can be both an ancestor and a descendant of another element), which I believe is possible in an ontology. Perhaps I’m mistaken.

  5. Rusty Alderson says:

    Mark,

    You are correct that an ontology is not hierarchical. An ontology, however, often “informs” the taxonomy, which _is_ hierarchical. Taxonomies and ontologies often work hand-in-hand. An ontology is the vocabulary for a given information domain, and provides the meat (definitions, relations, constraints, rules) for the taxonomy’s skeleton.

  6. Ariel Lee says:

    Mark, thank you so much for this very informative article. I’m currently working on refining the categories of our online store. (All in Korean unfortunately) I pretty much agree with everything that you’ve said about implementing the idea of ” ontology” for improvement but i am actually having trouble persuading the board members that this is the right move.
    Also it’s very difficult to draw the line between the two.
    What should be heirarchical and what should be non-heirarchical?

    For example, currently at our store, “Roundneck” goes underneath “Clothing > Women > T-shirt” in the hierarchical format but i think that “Roundneck” should be given in the non-hierarchical format – filters with checkboxes given.

    If you have any comments on how to define the two terms please share.

    Cheers,
    Ariel Lee

OfficeFolders theme by Themocracy