Which service types can't you classify against the ESDService list?

Technical
,
1

Hi all,

I’ve been working at iStandUK on automatically classifying ORUK service feeds against the ESD Service ValueSet (210 concepts) and I’d value a sanity check from anyone else who has tried this.

We classified 6,890 services from three published v3 feeds - Bristol, Shropshire and North Lincolnshire and then sat three of us down to hand-classify a sample properly with the full description and the closest matching concepts in front of us.

We got stuck faster than expected and not for the reason we assumed. The right concept was usually among the options we were shown. The problem was that the concepts available are broad enough that very different services end up in the same bucket. A community choir, a monthly women’s social group and a self-help group for a long-term condition all land in Social clubs and groups or Health support groups and organisations, because nothing more specific exists. Across the whole corpus one concept absorbed 994 services and the five most-used concepts account for about 30% of all the codes we assigned.

A few service types we couldn’t express at all: charity shops, community safety schemes like Safe Places, veterans support, and services defined by a specific condition such as dementia or autism rather than by the kind of help offered.

My sense is that ESD was designed around council service functions and we’re trying to use it to describe the voluntary and community sector, which is most of what these directories actually contain.

So, three questions:

  1. Has anyone else classified ORUK service data against ESD (or any other taxonomy) and did you hit the same thing?
  2. Which service types have you been unable to classify? If several of us are hitting the same gaps, that’s evidence for a proposal rather than one team’s opinion.
  3. Does anyone hold service data that is already classified, against ESD or anything else? Ground truth is the thing we can’t produce ourselves and even a few hundred labelled services would let us measure properly rather than guess.

Happy to share what we’ve built, the classifications, the method and the numbers with anyone who finds it useful.

Thanks,
Jay Prajapati
iStandUK

2

Welcome to the taxonomy problem, i’m sorry to inform you there is no way out. but at least you are in good company.