Search on eBay for “Sudbury Hall” in collectables, as a phrase. It’s the National Trust Museum of Childhood, some 12 miles east of Stoke-on-Trent on the Staffordshire / Derbyshire border.

eBay’s ‘Recommendations’…

What a load of rubbish. So much for the much-hyped advanced in semantic search capabilities and sophisticated tailoring of search to user data.

This is on eBay, but it’s just as bad on Amazon. And Pinterest. Search Pinterest for staffordshire postcard -dog and get…

But I just told Pinterest (-dog mean no posts which mention the word dog) I didn’t want any dogs, so why the hell is Pinterest still recommending dog stuff to me? Grrr.

Search recommendation systems are obviously running on pathetically broad and isolated keywords. It’s even infecting pure search. For instance, Google Images seems to be rapidly becoming so fuzzy in the relevancy ranking of its results as to be unusable. Why can’t huge billion-dollar world-leading tech companies get this right?

A proper system for eBay might be something along the lines of:

  User is logged in – yes.
  Where is the user known to be based? UK, Midlands.
  Is the search phrase a recognised placename – yes.
  Compare current search phrase to user’s search history. For this type of search the user is expecting results from Staffordshire, Derbyshire, Cheshire.
  Does the place name geo-match with the user’s search history and known location – yes.
  Therefore – cluster all sidebar recommendations on the UK Midlands, and exclude listings from all other UK areas.
  Then filter Midlands suggestions so that they only show ‘known place’ items + cluster within a fifteen mile radius of “Sudbury Hall”.

How difficult can it be? And how much would user-goodwill and profits be boosted, when they stop showing totally irrelevant suggestions?