Re-tune work search ranking parameters

[cdrini]

We've made a number of changes to our search, and it's time to re-evaluate / re-tune some of our search parameters via the evaluation spreadsheet.

Potential tweaks:

1. we should decrease the edition_count boosting: bf='min(100,edition_count) min(100,def(readinglog_count,0))' . I think this served as well while we were still getting more readinglog_count usage, but now that we do have a lot of readinlog_count, I think it's causing non-relevant classics to get boosted. We should cap it closer to maybe 20, if that.
2. We might want to increase the phrase boosting, pf. I'm wary of adding the pf2/etc parameters, since I'm afraid they might have a perf impact, but worth experimenting with.
3. We explicitly specify AND ; but having a mm parameter, which allows long queries to only partially match ~75% might be interesting! But I don't think that's what the data is showing is the problem.
4. Stopwords. We had to disable stopwords many years ago due to a bug solr had ; it would be worth revisiting that to see if we can re-enable them. Note this would require a full reindex.
5. We have a really annoying subtle/low-level issue with the way search works related to term frequency and merging ; that would require a non-trivial solr config change/investigation. See Investigate peculiar search ranking #10969 .

Note this will require exposing some low-level search parameters for us to play with on testing. Eg #8482 .

Stakeholders

@lokesh

[cdrini]

Some examples of some odd results from @samar-703 in #11694 :

• Search: Thinking Fast and Slow; Shows the actual book by Daniel Kahneman in 4th place.
• Search: Skin in the game: Shows an unmerged work first for some reason. The borrowable edition with higher engagement should rank first

[samar-703]

Example: Rich Dad Poor Dad

• Search: Rich Dad Poor Dad: Shows the sequel ranking 1 while the original book, with significantly higher engagement, better ratings, and more editions, ranks second.

[cdrini]

Thank you @samar-703 , I added to our evaluation spreadsheet. Please provide any more examples you find and I'll add them in!

[lokesh]

First result in search on the test server, huge improvement! Both the book search issues I previously reported are now returning 'correct' results.

I've added another book to the evaluation that I've struggled to find at the bottom of the evaluation doc: "the work of art"