ia play

I am a bit obsessed by ‘when recommendations go wrong’ scenarios like the JustRabbitHutches incident.

iPlayer hasn’t done anything that silly, but it does seem to struggle with the recommendation concept. They sit particularly uneasily alongside the new Favourites functionality.

When the latest incarnation was in beta, I was quite excited by the prospect of favourite programmes and categories functionality. This had the potential to meet some of the needs that the absence of sophisticated browse function left. If I could tailor the content more then I’d need to browse less.

But the new site makes surprisingly little use of the favourites functionality. After you’ve put the effort into setting your favourites, it pretty much ignores all the work you’ve put in.

The favourite programmes bar is always closed. The favourite categories are similarly always closed. The radio stations box doesn’t remember your selection.

The homepage is dominated by four sections: Featured; For You; Most Popular; and Friends. None of these areas seem to be influenced by your own preferences.

Featured is rarely of interest to me but I get the editorial need to have some promo space.

For You is where the recommendations kick in but at least initially I had no idea what this section was supposed to be doing. A good design pattern is to explain recommendations ala Amazon and to let you know if there is anything you can do to make the suggestions better.

Most Popular is ok for me. Occasionally my interests overlap with the majority and then this spot is useful. Friends might be occasionally interesting, although “a people like you like” might have been more valuable. It seems a bit odd for the area to persist if you don’t login/specify any friends.

All these sections are potentially useful but the best predictor of my interests is my interests. It seems that in this design My Favourites and My Categories are given lower emphasis than *everything* else.

This is compounded by the presence of the For You section. As another commentator put it:

“why on earth would the site suggest I watch Eastenders? It’s been on TV for over 25 years and I’ve never once felt inclined to watch it, so what intuitive masterstroke has been developed to think that I may now wish to start?”.

Once you give recommendations personal labels like “For You” then people start to take your recommendations personally.

I’m annoyed that I told iPlayer what I like and it still insists on telling me that BBC 3 sitcoms are “for you!”. It’s started reminding me of my grandad and that’s not a flattering comparison.

My rabbit hutch purchasing has been an interesting vein of UX experiences. In the end I bought a hutch from JustRabbitHutches, whose website was mostly pleasant to use and whose service was great.

That said, once I’d added my hutch to the basket I noticed they’d been tripped up by recommendations. Under my basket were suggestions that I might enjoy. Unfortunately one of them was a “delivery surcharge”.

Now this isn’t as damaging as Walmart’s dodgy DVD recommendations but it’s another example of how careful you have to be.

You could also ask why JustRabbitHutches thought they needed a recommendation engine here. After all the clue is in the title. If I’m buying a rabbit hutch, how likely is it that they’ll be able to sell me another one?

I always like examples of recommendation engines and the like that have got a bit muddled. The WalMart Apes scandal remains the classic. In this case the book is Apocalypses: Prophecies, Cults and Millennial Beliefs Throughout the Ages and the sponsored link reads “Cheap Weber BBQs”.

It would be nice to think that the suggestion that customer interested in a book on apocalypses might also like a BBQ had some sort of ‘burn in hell’ connnection but it appears to just be that the author is called “Weber” which is a BBQ brand.

Which started me thinking about how to improve the recommendation engine with a bit of semantic insight about which fields to match upon. You could just not match on the author field but presumably some of the sponsored links are actually related to the author (I’m thinking the Gillian McKeiths and Deepak Chopras of the world). So you’d need some semantic information about the content of the sponsored link as well. Which could be a bit more challenging…