Targeted ads have a long way to go in the semantic web

c: | f: /

Personalised banner advertising on the Internet is an utter waste of money, effort and bandwidth because the industry have made some fundamentally wrong decisions about what constitutes useful metadata.

I had a job interview as a web developer at a training shoe company a few months ago. Yes, really. In the interests of research I visited their website and clicked on a pair of trainers to see the type of data they stored against each product. At about the same time, I was shopping for a Christmas present and clicked on a jewellery box on a site (I think it was Marks & Spencers).

Every one-in-three, or one-in-four sites since then — presumably ones that share a media server — show me a sidebar advert for exactly the same jewellery box and exactly the same pair of trainers. After just one click on each item, my shopping habits have been analysed and distilled. And are plain wrong.

Only after about three months did the adverts wane and revert to the usual crop of random ads that I routinely ignore. I’m mostly advert blind, and the only reason I noticed the ads at all was because they exactly matched the products I clicked and I thought it was equal parts clever and scary that the advertisers would pay someone to do that to my browsing experience.

Then I got thinking about this whole targeted advert paradigm. As the web and search engines march towards the semantic search nirvana of only advertising things that I truly want, and only returning search results based on things in which they think I’m interested, a lot of people are going to be disappointed by the reduction in choice.

A good analogy is this: imagine walking into a supermarket stocked floor to ceiling with every product you could ever wish. Like Wallis and Gromit, you have a hankering for a tasty block of Wensleydale cheese. But the moment you cross the threshold of the shop, 95% of the products vanish from the shelves because the supermarket only show you what you’ve bought before and, wrongly, assume you only want to buy the same things this week, which includes cheddar cheese.

That’s targeted advertising.

Based on my prior habits and customer profile, the supermarket have “enhanced my shopping experience” so I can more easily find that in which I’m routinely interested without wading through the other products I’ve never bought. Well, to me, that erodes choice. I’d much rather make the decision than have some piece of software limit what I might be thinking and force me to go out of my way to circumvent its decisions.

In a similar vein, Amazon gleefully send me spam messages giving me “recommendations” for things I might want to buy based on products I’ve bought in the past. It seems to have passed them by that 65% of the things I buy on Amazon are an eclectic mix of items for other people as gifts. Sure I could go in and battle their software to tick all the things I actually bought for myself and exclude gifts so they can make a better guess at what I might want, but then it’d suggest things too similar to items I already own, which defeats the purpose of them sending the messages to try and entice me to buy stuff I don’t need.

Search engines are doing the same. I get offered a list of cars when I search for “jaguar” because I’m a bloke and must therefore be interested in four-wheeled vehicles (regardless that I’m not). To shunt the four-legged wild cat — the actual target of my search — to fifteenth place (which is on page two of most people’s results), is pure and simple commercialism: the sell first, everything else second mentality. In a true semantic search, the results would be mixed or would allow me to disambiguate up front rather than make the wrong assumption and frustrating me. Google Images allows this to some degree for some search terms, but there’s a long way to go before it’s truly useful.

In its current guise, targeted advertising and semantic search are simply broken. And I’m confident in saying that while the corporations are calling the shots, they will remain so.

Your two cents


(required, never made visible)

(optional, linked with rel="nofollow")