How does Trove's relevance ranking work?
The relevance ranking is complex and takes many things into account.
- A record with a match in the title, subject or author field is considered more relevant than one with a match anywhere else. (But this is complicated by the number of matches found, and other rules which can push records higher or lower)
- Exact matches are ranked higher than not-quite-exact (stemmed) matches. (e.g. You search for Jack, we match Jackes. Items that don’t have the exact word you used your search are not as relevant as those which do.)
- Search terms close together are ranked higher than search terms far apart.
- If two records have very similar titles, but the second record also has the search terms found in the notes or full text of the record, the second record will be considered to be slightly more relevant than the first.
- A record which has a higher “density" of matches with the search terms is ranked higher than a record with a low density. (That is, a short sentence with one occurrence of your search terms will rank higher than 100 pages of text with one occurrence of your search terms)
- Items where your search terms are only found in the full text (or in the finding aid) are not as relevant as items where your search terms are found in the title, author, subject, note, etc.
- Items from "my libraries" or online are more relevant. Items with no holdings or other getting options are less relevant.
- Search terms found in a single work but split across different editions will be ranked very low
Why do the decade facets show wrong years?
The reason that these wrong years are appearing is due to the way that the data has been encoded by the organisations that have provided these records to Trove.
Trove accepts records in a variety of different formats, and most of these formats include a specific date field which is used to populate the date facet in Trove. With some records, the date is not supplied in a standardised format, and Trove tries to interpret the information in the date field to produce a date for the record. Unfortunately, in some cases these dates are interpreted incorrectly, and produce decade ranges that are completely wrong.
Investigation into a solution is currently underway.