Observations from a reconciliation process

[susannaanas]

Describe the bug
I created a memo of problems with a reconciliation process and enhancement ideas. I will break it down to proper issues after finishing the project and possible comments, corrections and removals.

Desktop (please complete the following information):

• OS: MacOS 10.13.6
• Browser Version: Version 75.0.3770.100 (Official Build) (64-bit)
• JRE or JDK Version:Java(TM) SE Runtime Environment (build 1.8.0_111-b14)

OpenRefine (please complete the following information):

• Version 3.2-SNAPSHOT [8d89a2a]

Datasets
https://drive.google.com/open?id=1_i-Gy2Q584I9V0ktiGdp2ME9qt5vvedk

Additional context

Updating OR in the reconciliation process
When a Wikidata import needed to be stopped to make a correction, the OpenRefine database and Wikidata got out of sync. The information already added to Wikidata was not updated to the OpenRefine items that had already been uploaded. On another occasion, they did not get updated although the update ran normally. To my knowledge, the new QID should be matched to the OR item, but if that is not the case, updating reconciliation data would be an enhancement request. The new items are however created in Wikidata and this will result in a new matching process.

After this interrupted Wikidata upload I was able to identify the record number where the upload has stopped. I would have liked to select records based on the record number range to correct them.

Matching by external IDs
It should be possible to match more explicitly based on an external identifier. It is ok for some cases that there is a fallback mechanism for a regular match, but that should be possible to switch off/control. It is difficult to know which of the matches are certain, and which are good guesses. This could also be solved by indicating the type of match in the UI.

The type matching mechanism could be repurposed to serve this need. One could select any property, such as subclass of, or an external ID, against which to reconcile the cell value which is not necessarily a label.

Popups
The popup matching interface is somewhat limited. If the item is not in the first set of suggestions in the flyout, there is no way to retrieve more.

Selecting an option (the item to reconcile with, or any of the options) in the popup makes the interface jump to the first page of the table. The checkboxes do not refresh the page in a similar manner. This remarkably slows down work. I know this is a known bug and I will give my thumb to correcting it.

The new hover interface forces to use the popup for matching and the page to refresh after matching. Perhaps the hovercard could be repositioned so that it would not cover the checkboxes.

The file gets slow over time (I wonder if it gets corrupted). The popups are impossible to use as they do not populate. It becomes impossible to reconcile. A workaround could be that much information of the reconciliation options that are now only shown in the popups dynamically could be imported to the file in the data retrieval phase.

Multiple rows in a record

Key columns with values in multiple rows causes misery and sorrow. It is unclear how records with multiple rows should be handled.
1 The key column needs to be filled down to be able to add the multiple properties to the item (I think). If any of the values for properties is declared within the schema rather than in the data, it is added to all the split rows, multiple times to one item.
2 When using the “Choose new match” popup for such a filled-down item with multiple rows, it is unclear if a new value is created multiple times for each row in the record, once to the whole record or multiple times to all records containing the same value or multiple times to all rows containing the same value. It is not displayed in any way after adding to be able to track errors.

It is unclear how to manage new items as values for properties in the schema. How to give the new valueitems necessary properties to be able to use them as values in the schema. It seems it should be made an additional schemaitem in the schema, but in which order is data read and created is not clear. This has resulted in errors, but I need to dig into archives to find the specifics.

→ It is unclear how to manage multiple items in one import schema.

Miscellaneous

There could be visual interfaces for geolocations. Maybe even the distance between the compared geocoordinates could be retrieved rather than the mysterious score.

[wetneb]

Many thanks for this very detailed feedback, it is very useful! I will try to reply with more detailed answers and links to existing issues later.

[wetneb]

Updating OR in the reconciliation process

Yes, reconciliation results can get out of date and this can be an issue. I do not really see what we could do against this problem: the current solution is to re-run reconciliation with the same configuration, how could this be improved?

Concerning errors during editing, we have #1782 where we need to figure out the right design.

To create a numeric facet on the Qids, you can create a "Custom numeric facet" and use an expression such as cell.recon.match.id.substring(1).toNumber().

Matching by external IDs

These suggestions are extremely interesting and I will post them on the mailing list of the W3C Entity Reconciliation Community Group (that you are welcome to join if you like!)

Popups

As you know we have #33 already but all the other problems you mention are worth having issues on their own I would say.

Multiple rows in a record

Again the problems you mention would definitely be worth having individual issues.

Miscellaneous

There could be visual interfaces for geolocations.

I guess this would mean adding some sort of maps in the item previews? If so then that belongs to https://github.com/wetneb/openrefine-wikibase.

Maybe even the distance between the compared geocoordinates could be retrieved rather than the mysterious score.

That is typically something I want to push for in the W3C group, following my recommendations in https://arxiv.org/pdf/1906.08092

[tfmorris]

This is good feedback, but needs to be split into multiple actionable issues for things to be improved. I'm not sure how many. Some of the complaints I recognize as duplicating existing issues, e.g. the 8 year old #33, but someone needs to go through them all one by one.

[thadguidry]

@tfmorris Agree, so perhaps label this as "to be reviewed" for whenever someone has that time later.

[tfmorris]

Label added, but feel free to do that yourself in the future.

[pyrog]

Matching by external IDs
It should be possible to match more explicitly based on an external identifier.
👍

Also, matching against properties seem useful too.
See reconciliation-api/specs#27

In World Port Index, the port have a ISO country code.
Matching it against country (Q6256) don't work, but matching it against ISO 3166-1 alpha-2 code (P297) will find all countries.

An new option "Reconcile against property" could show the column in the right pane, so we could "include it as property"
P297.

[pyrog]

Miscellaneous
There could be visual interfaces for geolocations.

👍 Could display an OpenStreetMap map, but at which zoom level ?

[pyrog]

Miscellaneous
Maybe even the distance between the compared geocoordinates could be retrieved rather than the mysterious score.

👍 Could compute a simple Geographical distance with a distance parameter setting