Taxonomies, Porn and Online Advertising

Categorisation mistakes can be embarrassing. In my book I cite a Walmart case where an auto-classification error made “racist” suggestions to customers (eg a search for Charlie and the Chocolate Factory turned up suggestions for movies with African American themes). (Actually, auto-categorisation software is not so much racist, as insensitive to very human contexts, histories and sensibilities).

Now Singapore’s very own SingTel Digital Media has been caught out inadvertently posting ads for its very respectable InSing portal onto a pornographic website. You have to know Singapore (and the culture of government linked companies like SingTel) to appreciate the furore this aroused here. Now I say, why shouldn’t the propriety of InSing be promoted among the heathen pornographers? And what were those Singaporeans who discovered the ad doing on that site in the first place (the “Disgusted” rapporteur claimed to have received the screenshot in an email from a friend)? And why do Singaporeans call porn “prawn” when posting online?

Turns out that the mistake was a categorisation error. The digital advertising contract was outsourced to .Fox Networks, with strict instructions on the categories of website for the ads to be placed on. .Fox Networks relies on third party providers to supply websites to advertise on, and one of its outsourced providers had apparently misclassified an adult website, leading .Fox to assume that it was worthy of SingTel’s money.

There are a couple of lessons here: (a) you can expect categorisation mistakes when you sub-sub-sub-contract (out-out-out-source) categorisation decisions (your common ground diminishes as you distance yourself from your categorisers); (b) categorisation errors can come back to bite you in the ass in very unexpected ways; (c) online advertising is especially vulnerable because it’s hugely expensive to audit categorisation accuracy – as in the Walmart case, you only discover the error when you create outrage. And that’s too late.

Yesterday’s post about the nasty consequences of misclassifying skate was about the dangers of rendering salient things invisible. The Walmart and InSing stories are also about invisibility: in rendering the categorisation process opaque, either through automation or outsourcing. Managers’ nonchalence about the rigour of their taxonomy regimes is a risky approach to take.

2 Comments so far

Tim Wieringa

Thank you very much for this post and the discussion on invisibility. For me the lesson (and confirmation for my opinion) is that when managing taxonomies, or actually any kind of automated/outsourced process, you should always know what you are doing. Don’t rely on a ‘black box’ and hope.
We are experimenting with semantics on TREX and my experience is: a) very interesting suggestions; b) not possible to release it unfiltered.

Posted on December 15, 2009 at 03:46 PM | Comment permalink

Tim Wieringa

TREX is a search engine run by SAP. I would not really recommend it and we are just trying to follow our IT policy.
The semantic engine of TREX is very experimental and we are still trying to get some documentation on it.

Posted on December 15, 2009 at 04:33 PM | Comment permalink

Page 1 of 1 pages

Comment Guidelines: Basic XHTML is allowed (<strong>, <em>, <a>) Line breaks and paragraphs are automatically generated. URLs are automatically converted into links.