The de-duplication should help as it brings together full and minimal records. Users will find the minimal records by virtue of us being able to associate them with the metadata from the fuller records. So a poorer record may be pulled up the rankings because it is part of a de-duplicated group of records.

I think it unlikely we can do much with our really poor records (those with just a title and little else.) It may not be possible to allocate them to a FRBR work level record and they may just have to sit alone and unloved in the database.

The current deduplication process is actually already a two-part match process so the move to having both work level and manifestation level records won’t actually add to the deduplication burden.

With the deduplication process it would be too slow to match every incoming record with the entire database, so we do an initial quick and dirty match to create a set of potential duplicate records. The incoming record is then matched against the potential duplicates using a second, much more detailed, duplicate checking process to confirm or reject the initial match.

We intend to refine the initial match procedure to bring together records for all the different manifestations of a work. This work level record set will then act as the set of potential duplicates for the second match stage, where we merge just those records for a specific manifestation of the work eg. the records relating to one particular edition.

So the match procedures will need to be changed but the overall work involved in matching should be much the same as now.

However I think Hugh is right in suggesting that work level record matching may result in a higher level of misconsolidation of records. The broader the match the more likely it is that we will bring records together inappropriately. There are still some ‘patient researchers’ out there who get in touch to report errors in the records, something we are planning to make easier to do. And we will put in place ways of dealing with reported consolidation errors to stop them happening again. But as Hugh says – ‘A tricky one’ and the diversity of the records is always going to make it difficult to resolve.

I’m slightly less clear re deduplication. I don’t think having a work level and an expression/manifestation level – interesting as that is in itself – quite addresses what I perceive to be the gist of the comments you refer to at the beginning of that para. The real issue here seems to be about the matching algorithm(s) – whether this is the same as that. By associating all these expression/manifestation records not only with each other but with a “work” you are adding to the deduplication. The fact that the patient researcher will eventually be able to determine that some of these records are inappropriately linked may be a bonus (if there are any patient researchers left out there…), but it means there’s likely to be more such knots to untangle in the first place. A tricky one which I don;t envy you trying to tackle!