Future improvements for 'similar bills'

[aih]

For use of shingles (multi-word phrases) with 'more-like-this', see, e.g. https://discuss.elastic.co/t/more-like-this-and-shingles-phrases/100775

I got it to work by combining multiple More Like This queries each with their own analyzer instead of trying to use per_field_analyzer. That worked out better anyway, allowing me to have separate settings (e.g. stop_words, min_word_length) for unigrams vs bigrams.

===
Flatgov discussion

From Daniel:

I think multi word phrases probably makes a lot of sense here than looking at single word phrases.

We can also probably limit small bill to bill comparisons based on how CRS/LOC categorizes them. There's only a handful of monster bills and those are the ones it's probably useful to identify when smaller bills are components. We can also potentially use the section headings as a clue to narrow that down.

We could also use the Library of Congress summaries. They are much smaller but have to identify the key concepts.... could be a way to winnow it down.

[aih]

n-grams are now implemented and working.