The Australian Government coat of Arms

Communities of practice

Communities of practice

Reccomendation engine and components - BOM?

Hello again. Is anyone here aware of the work the Bureau of Meteorology did on what I recall to be a cross-government recommendation engine? It was a little like a component that we were going to be able to insert on sites that would show users similar messages or pages from across government based on the implicit information we knew of the user. Very similar to how Adwords work.

Sounds interesting Lisa, I hadn’t personally heard about that before. Do you have any code, links or papers to share?
I know the Observatory team has a similar concept in their backlog and would be very interested.

@pdb

Hi Lisa,

I am unaware of BoM’s work on this, but we do have a backlog story looking at using the whole of government webcrawl to generate a recommendation engine. The blocker we are still looking at is categorising the various pages to build the train set, which would take a significant amount of time.

Patrick

UNOFFICIAL

I’m afraid I have nothing other than the memory of a discussion about it and perhaps seeing a demo of it in maybe 2015. I think it arose out of the BoM work to get permission to host ads on its website.

UNOFFICIAL

I’m sure I read recently that Gov.uk are looking at using AI to assist with categorising. Unfortunately I can’t find that information either. Another option we looked at a while ago was the idea of user tags to assist. If you could create an app or function that allowed users to group their government information under their own titles (say like StumbleUpon used to or Pinterest does now) and there was a benefit to them it could speed up the work. We’d need to be able to curtail the ability for it to go off the rails. Maybe we create our own ReCaptcha and get people completing form to demonstrate they’re not a robot by categorising a few web pages for us :wink:

Hi Lisa,

I think this is what you are talking https://dataingovernment.blog.gov.uk/2018/10/19/how-we-used-deep-learning-to-structure-gov-uks-content/. This work had the benefit of pre-built train set, but this work showed we are on the right track.

PDB

UNOFFICIAL

Would it be possible to analyse Twitter data to see what hashtags people have used when posting .gov.au links?

Yes, but ideally you would want a variety of people categorise webpages too have a better spread for your train set. And it wouldn’t necessarily reflect other types of pages that have limited interest, especially those providing information used by other public servants.

Also, unless you pay for access to the Twitter API, otherwise you receive a sampled response that changes from search to search.