AI in Moliri
Finding duplicates
Duplicates come about easily: an object recorded twice, an artist in three spellings, a person who came into the authority data from two sources. Moliri looks for them in several places. Almost all of them compare the values themselves and need no model; a language model helps where values mean the same but are written quite differently. Nothing is merged or unified until you click.
Overview
| What is duplicated | Where Moliri finds it | With a model | What you can do |
|---|---|---|---|
| An object recorded twice | Suspected duplicates under Check | No | Show both in the gallery, clean up one or choose Not a duplicate |
| A value that should occur only once, such as an inventory number | Duplicate values under Field conditions | No | Filter the objects concerned and correct them |
| A row of an Excel list whose object already exists | Matching in the Excel import | No | Assign the row to the existing object instead of creating a new one |
| A value in several spellings, such as Öl a. Lw. and Öl auf Leinwand | Similar values under Check | No | Set all of them to one spelling |
| A value with the same meaning in other words, such as Florenz and Firenze | AI run under Similar values | Yes | As with the spellings |
| Two authority records for the same person or place | Syncing with the source and Similar values, Authority data tab | No | Merge |
| A literature reference that is already in the project | New from reference | No | Open the existing record or create one anyway |
Duplicate objects
Suspected duplicates compares the fields on which the administration has switched on Suspected duplicates, across the whole project. It ignores upper and lower case, accents, punctuation, spaces and leading zeros: 1921.1239 and 1921/1239 count as the same. A pair that is not a duplicate is remembered for all projects with Not a duplicate. What the dialog looks like is described under Suspected duplicates ; how the administration sets the switch, under Fields .
Spellings and meanings
Similar values first finds, without a model, what differs only in how it is written: umlauts, typos, word order and abbreviations. Müller, Hans and Hans Müller end up in one group this way.
What has a different name but means the same is found only by the AI run: Florenz and Firenze, Aquarell and Wasserfarbe (watercolour). An embedding model looks for pairs whose meanings lie close together, and the chat model decides whether they really are the same. The groups appear at the top of the list with the badge AI suggestion and the model’s reasoning, and are handled like the others. The AI run needs an AI account with an embedding model and runs as a job on the server, every night if you wish; see AI jobs and nightly runs .
Duplicate authority data
The sync with the source reports Duplicate when another record already holds the identifier of a record, such as a GND number. The authority library shows such pairs with the Duplicates selection, and for the administration also under Check, Duplicate authority records. Records with similar names but no shared identifier are found by Similar values on the Authority data tab. Both can be merged from status 6, with a side-by-side comparison, a choice of direction and a dry run; afterwards the objects point to the record that remains. The steps are in the guide Merging duplicates .
When importing and creating
The Excel import finds existing objects through a column, usually the inventory number, and adds to them in the Complete and Overwrite modes instead of creating new ones. If a number that should be unique is duplicated after the import, the report names the row; this applies to duplicates within the table and to duplicates of the holdings.
In a literature project, New from reference warns with Already in the project when the same record from the source already has an entry, and offers to open it.
Further reading
- Check : all checks of the tab
- Authority data : sync and merge in detail
- What the AI can do : the other AI functions