multi-word exact matches at the start of a title aren't always first in the list of search results

[harrisonpim]

Hi gang, search ranking question - I was trying to find the book "Information Law", by searching on those words: https://wellcomecollection.org/works?query=information+law. As you can see the book doesn't show up until about half way down the second page of hits, even though the words are an exact match to the first two words of the title. Is that to be expected (because the book's works page isn't very popular or something?). If I add the author name to the search, then it shows up as the only hit (the book is https://wellcomecollection.org/works/zkg7xqm7).

[harrisonpim]

the current query seems to be giving high scores to matches in the search.relations, which has no positional awareness. we don't currently prioritise proximity to the start of a title (or any other field)

[harrisonpim]

The 'correct' result for this query can be retrieved in first position by adding a span_first query:

{
  "query": {
    "bool": {
      "should": [
        {
          "span_first": {
            "match": {
              "span_term": {
                "data.title.shingles": "information law"
              }
            },
            "end":1,
            "boost":1000
          }
        },
        {
          "multi_match": {
...

[harrisonpim]

I'd like to investigate the effect of

• turning down the boost on the relations field vs
• adding the span_first to title.shingles
  using the standard set of rank tools before making any changes to the prod query

that requires:

• CCRing the prod works index into the rank cluster
• adding a new precision test to the rank suite using this example
• running the rank test suite from CLI on both candidates
• running the rank speed comparison on both candidates

[pollecuttn]

Issue first reported on Slack at https://wellcome.slack.com/archives/C8X9YKM5X/p1653293886888919