If you can’t do an exact phrase search in just the headings, you could always add a quick check into the harvester (or whatever post-processing after the harvester has run) to do a check for the exact phrase using a regular expression – I’ve had to resort to this in the past when dealing with imprecise search interfaces.

I don’t know if any of this is helpful but I had some questions/thoughts as I read the above:

A) I’ve tried to recreate your search but get different results depending on the approach I take – and I’ve not yet managed to recreate your exact search!

For example I note that this search http://bit.ly/tphXME gets 294 hits (less than yours), but includes for example the editorial from the 2nd January which your search omits for some reason.

I also found variance depending on whether I used the date range in the advanced search, specified the data via the faceted filtering

B) Length of article looks like it might be another possible filter criteria – it seems on relatively brief inspection that editorials are longer than articles

C) How do you tell an article is actually an Editorial if reading the paper? It seems impossible to actually tell systematically? (is it just a matter of ‘guessing’ when you read it?)

D) I’m not familiar with how editorials would be structured, but I notice that there can be multiple ‘sections’ within an Editorial column with a single heading. For example this editorial http://trove.nla.gov.au/ndp/del/article/15388548 is headed ‘Traffic Growth’, but that is actually only the first part of the article. There is then a double rule, followed by (I think) 11 ‘sketches’ – short paras on different topics which are headed only by a short title followed by an em-dash (I think – could be en-dash?). Not sure how much this matters once you get into the textual analysis stage – but it seems a shame these aren’t separated out into articles.

Just dumping some thoughts really – sorry if not helpful!