13 June 2017, by Michael Kearns
Full Fact are a British independent factchecking charity who check claims made in the press, in parliament and in televised debates (e.g. Question Time). They’re currently in the process of implementing tools to enable automated factchecking (note that they’re not looking to do this in a machine-learning sense of “automated”, but mainly building tools to enable humans to check facts significantly faster).
They recently ran a hackday for the first time, which I volunteered at. They had a few problems which they wanted to tackle via the hackday:
- Wrapping a reverse-search library in a service, so that they can integrate it with their systems (and use it without learning Java, they mainly use Python)
- Implementing a way of finding claims of the form “X is rising” (e.g. “GDP increased by 5%”)
- Implementing a way of canonicalising numerical parts of speech (e.g. “three thousand and fifty” goes to “3050”; also needs to handle things like “few thousand”) so that claims made e.g. in Parliament are in a suitable format to potentially be automatically checked against an appropriate source.
I ended up volunteering for the first task, which in retrospect was definitely the least challenging (so I didn’t learn much) but was probably where I was most useful for Full Fact.
The hackday had quite a wide range of participants, from committers to apache-solr to developers inexperienced with the Solr, along with a special guest representative from the Argentinian fact-checking organisation Chequeado. Good progress was made on all three tasks, although I definitely feel there was a case of too many cooks on our task, along with some various set-up issues which caused the day to be quite inefficient for some volunteers (Java versions, IDE issues etc.)
I gave them feedback on these issues – this was the first hackday Full Fact have run, and they’re now looking to make use of everything that’s come out of it before coming up with lots more problems they need help solving for (potentially) a next one!