A few years ago I wrote about some metadata that was 4,000 years old. Today I wanted to write about another high point in the history of metadata — well, some sort of point: the failure of Amazon’s folksonomy and the Playing with Fire album by Kevin Federline, the former Mr. Britney Spears. Throughout this discussion I’ll show some of the tags that Amazon users added to this album, and I’ll also provide a link to a page where you can see the top 100 entries. (As you’ll see from the link in the previous sentence, these tags are no longer on Amazon’s page for that album.)

worst album ever
an assault to decency
bird vomit
every track ought to be hidden

The Wikipedia page for “folksonomy” begins by defining it as a “a classification system in which end users apply public tags to online items, typically to make those items easier for themselves or others to find later. Over time, this can give rise to a classification system based on those tags and how often they are applied or searched for, in contrast to a taxonomic classification designed by the owners of the content and specified when it is published”. The rest of the page gives a lot of good background. (It also says that the “study of the structuring or classification of folksonomy is termed folksontology” which looks like good work going on, but I sure hope they come up with a better name.)

For another good definition, slide 6 of a presentation (pdf) by Christine Connors, who is currently at Raytheon but whose name will be familiar to anyone who has done much work with taxonomies and data modeling, has a nice summary of what folksonomies are: a “people’s classification management” system that draws on the wisdom of the crowd to apply user-generated tags to digital resources so that, as with any tagging system, people can find the resources they need more easily.

When folksonomies first became popular, some people became overly excited. For example, the 2007 paper Folksonomy: the New Way to Serendipity (pdf) claimed that “folksonomy allows various modalities of curious explorations: a cultural exploration and a social exploration.”

music to make you long for the sweet release of death
should be working at wendys
tiresome and vulgar
vanilla ice

A lot of professional taxonomists didn’t like folksonomies, because without some measure of control then you don’t really have a controlled vocabulary. You have a a free-for-all. Quality taxonomies are built around a governance process to ensure that terms are added, revised, or deleted in an organized way. This process also ensures that these terms are presented consistently, and the resulting metadata helps users find the resources they need more efficiently. I attended the Taxonomy Bootcamp conference for several years when folksonomies were hotter, and based on many presentations that I saw, it was clear that this hotness had many in the community feeling a greater need to prove the value of their profession.

The Wikipedia page includes a pretty good list of the advantages and disadvantages of folksonomies, but it omits one of the key perceived advantages at the time: free metadata. Metadata’s purpose is to add value to data by making it easier to navigate, and when Amazon decided to let anyone tag any product with anything, they thought they were getting people to add value for nothing. It was an early hint about where the eventual “creator economy” would go: people doing free work for a multi-billion dollar company for the sense of community and for the satisfaction in seeing their work on the big famous platform, but certainly not for any worthwhile amount of money.

What could go wrong? Let’s look at Exhibit A. Thanks to the Wayback Machine we can look at the January 15, 2008 version of Amazon’s page for Mr. Federline’s debut album. Let’s look at an excerpt:

Lots of interesting information here!

• The first tag every applied was “poser loser”
• The last tag was “stupid”
• The most popular tag was “talentless”, which 45 people applied
• The second most popular was “music to make you long for the sweet release of death”, which 27 people applied

And so on. As you can see, the tags in that list were links, but they no longer work. Back then, each of these linked to a page that listed everything else with that tag. The first one went to the excellent URL http://www.amazon.com/tag/talentless, which no longer has anything on it, but you can see a Wayback Machine version from that era that includes some of Paris Hilton’s work. At one point that older version also let you “narrow by popular tags” like “horrible” and “trash” as a sort of faceted search.

frisbee
i left justin for this
pure concentrated evil
cole slaw

Wikipedia’s Kevin Federline page says that the album is “commonly considered to be one of the worst albums ever released”. That links to the Wikipedia page titled List of music considered the worst, where it is the first entry under “2000s–2020s” (which, to be fair, is sorted chronologically). The album is not even on Spotify, but if you really want to hear an awful attempt at West Coast hiphop, it’s on YouTube.

Because of tags like these, Amazon eventually realized that folksonomies were not adding worthwhile value. They discontinued the use of folksonomy tags and eventually removed the About Tags page that explained how they once had this feature available but didn’t anymore. As it says, “We’ve since continued to innovate on our more popular features such as Wish Lists, Customer Reviews, and Improve Your Recommendations”. More innovations in the creator economy!

So taxonomists didn’t have to worry about proving the value of their profession. In fact, as more people learn that carefully curated knowledge graphs are a useful tool for reducing the number of hallucinations coming from Large Language Models, these people are appreciating that taxonomist skills are an excellent approach to curating those knowledge models. I’m sure there are still plenty of Content Management Systems in production where users can make up their own tags, but I’ll bet trained taxonomists are reviewing and normalizing those contributions as part of their job of improving the system’s metadata.

makes baby jesus cry
dumbass
rich wife
ear bleach

Comments? Reply to my Mastodon message or bluesky post announcing this blog entry.