Why keywords aren't meaningful communication in ONIX — BookNet Canada
, ,

Why keywords aren't meaningful communication in ONIX

, ,

The definition of quality in metadata is getting the results you want by trading for it and delivering what’s needed. The metadata needs trading partners who are also getting the results they want.

Do keywords really bring accurate results in a search? @BookNet_Canada's Tom Richardson goes over the use of keywords in ONIX and their limitations.
CLICK TO TWEET

I just heard someone mutter something about “when was the last time [their so-called trading partner] gave feedback? Like never.” Naming the problem is always the first step in its solution and this person named the problem. Therefore, we can see that the real definition of quality in metadata is meaningful communication. The supply chain is just a rom-com full of strife and missteps until the players realize that they both want the same thing and a contract and happily-ever-after (profit) ensues. 

If the real definition of quality in metadata is meaningful communication, I’ll try to demonstrate that quality as it solves business problems. I’m going to let you know how you’ll be rewarded if you dare to try. Take one use case as an example of unintended consequences in keywords: BNC CataList.

When my colleagues at CataList heard that my topic for Tech Forum 2020 was data quality, the first thing they said was “Ya gotta talk about keywords. It’s bad.” This is also something I’ve heard consistently from retailers. Keywords would seem to be something publishers provide to improve search. So what goes wrong at CataList?

Here are some examples of ineffective searches on CataList:

Both “Harry Potter” and “COVID” are now ineffective searches since you have to sort through pages of results until you get to the results you were likely looking for:

Author searches return books by comparable authors as well as books by the searched author:

Search results, therefore, need careful study to determine irrelevancy, an activity that offers both the cataloguer and the searcher low return on investment.

So: does it make any sense to supply your trading partners with keywords that are designed to manipulate consumer search results on a specific retail site? Or: can you manipulate Amazon’s search meaningfully given it has the world’s most developed and complex consumer data and that keyword creators can only access a small portion of it?

Retail search optimization may belong with the site owner, and not with their data suppliers, but I can confirm that search effectiveness degraded on CataList when keyword searching was added.

BookNet Canada’s sister organization the Book Industry Study Group released their Revised Best Practices for Keywords in Metadata in 2018. It contains a lot of good advice on keywording.

In the end, it’s not really my place to say how keywords should be developing but I think the focus on Amazon is troubling from a data quality perspective. And it seems that adding keywords is perceived as the new “all-in-one” solution, ie. if you can’t stick it in the title field then stick it in keywords. But, will your sales improve?

I don’t want to blame publishers for how they use keywords. They’re doing what they can to increase their sales. Data quality, when defined as meaningful communication, would require retailers being clear about what they index to support their search.

Good data quality should offer a reward and that brings me to the concept of “small data”: uniquely identified data points that are specific, directed, findable, and fully accessible.

That is, undoubtedly, an aspirational definition and not a practical one. Small data is when a specific data point appears on less than 10% of the books in an aggregated database but more than 90% of the books in that aggregate carry more than one of these small data points.

I’m critical of keywords but they would seem to represent the ultimate piece of “small data” as they’re uniquely chosen to represent the book. But the difference is that you don’t know why a keyword is there. It’s a mulch that can include references designed as clickbait squatting in the middle of a database. Your product metadata mixing with all the other choices made by less able firms. Small data, on the other hand, comes unique but identified. It always pertains directly to the book. It’s another way to say granular data. And it’s the antidote to the problem I’ve identified in the past: data producers focus on requests made by end users. Who uses what and how do they want it. While end users only interrogate the data they receive.

That’s a loop and it’s a loop that can only find data in volume. Data creation and use done in that way won’t find “small data” — the little bits that give distinction and meaning that actually drive discoverability. Which is what publishers seem to want from keywords!

Certain types of books have needs. For example: educational books come with different expectations. On gross terms we deal with these using Audience codes. Children, young adult, and the trade have different needs. But we dump them in the same aggregated metadata bin and ignore the differences, supporting each as much like the other as possible. Sure, we can be sure the “data is used” and that “the data will be supplied.” But we’re left to wonder why. Diversity, for example, is poorly supported. Supporting small data, granular data, is part of the answer.

Small data is diverse, unique, and distinguishes books from each other. Keywords are a poor substitute for small data. Thema, a structured predictable subject system of great nuance, would provide far better support for the use of small data. 

This post is an excerpt from BookNet Canada Bibliographic Manager Tom Richardson’s presentation for Tech Forum 2020. Watch his complete Tech Forum presentation here and download his slides here.

For more explorations on the use of keywords in product metadata, consider the following resources:

For more coverage on keywords, consult the BNC Blog.