Unwanted Dynamic Tagging

DEVONthink 4.1 Mac M1 Mini 16GB Sequoia 15.6.1

TLDR: How to remove unwanted dynamic tags when the relevant settings do not mention dynamic tags

In DEVONthink→Settings I have nothing active relating to tags:

In Database→Properties I have no tagging:

I have imported three folders of PDFs into a new database:

I created my Tags, placed them into categories and spent time color coordinating them

I then assigned tags to one document:

Feeling a sense of accomplishment, I prepared myself to apply tags to another file. But, I notice that it as seemingly already acquired some tags:

I can’t remove these tags (note that they do not have my color scheme). I can click on them and the node-map in the Related Tags window changes.

Looking at another imported PDF in the same folder, I see that it also has automagically acquired some tags:

Looking at the rest of the PDFs in this folder/group and they all have acquired tags. Not all the same tags though. Looking at the original files as they exist within Finder and there are no tags applied.

Reviewing PDFs in a separately imported group, I see that they too have acquired some tags:

When I go to the Tags and unfurl the categories to see the Tag itself, it shows that only one item is assigned to that tag (the one single item that I tagged manually). It does not show any of the other files that have magically acquired tags.

The Tag properties appear uneventful:

What is necessary to remove these dynamically applied (unwanted) tags.

Needless to say, it’s driving me a bit potty.

Cheers

dp

As a start, perhaps the tags were in the original PDFs before entering DEVONthink?

Just reenable the option Exclude groups from tagging in File > Database Properties. Currently each enclosing group is automatically a tag as it’s disabled.

Definitely not. I mentioned in my original post that I had confirmed the absence of tags at the Finder level.

Just to ensure that there were no tags on the Finder-level file I ran the following command from the terminal line:

xattr -d com.apple.metadata:_kMDItemUserTags {/fullpath/filename.ext}

Which returns:

No such xattr: com.apple.metadata:_kMDItemUserTags

Thereby proving that there are are no attached tags to the file.

1 Like

I re-enabled the option Exclude groups from tagging in File > Database Properties as advised, but this did not remove the superfluous/dynamic tags that were currently attached to each file.

I removed each file keeping the one that I had manually tagged, emptied the trash and reimported files singularly from Finder.

As each file is imported I am able to confirm that it has no tags.

However, as the DEVONthink Indexing Items activity is running I check the files again and I can see that gradually each file is getting Tags automatically/dynamically attached.

To further confound the problem, there are different sets of Tags dynamically assigned to each file. I can see that the majority are my Parent Tags (or top-level Tags from my nested Tags) but many are child-level Tags. But neither the presence of Parent Tags or Child Tags is consistent across the files.

As I write this, the Indexing Items process is now complete and the newly imported files have anything from 1 to 10 dynamically assigned Tags.

If the dynamically assigned Tags were correct I might not mind, but there are several instances where it is just clearly wrong. Some files have not yet undergone the OCR process and yet these are also receiving dynamic tags.

Also, I’ve just noticed that these dynamic tags only appear in the left-hand side bar (the Inspector?) and not at the bottom of the file when the Show/Hide Tags from the menu bar is active.

Any guidance is appreciated

Cheers

dp

The shown tags are just suggestions provided by this inspector. The assigned tags can be viewed e.g. via the Info > Generic inspector or the Tags bar at the bottom of windows.

1 Like

Further testing has demonstrated that the dynamic Tags are automatically applied when document indexing occurs and produces the Word Cloud and Word List.

From what I can gather document indexing “reads” through the document content for words that match existing Tags and dynamically applies them.

However, while the word may indeed appear within the document there are a few issues:

  1. the context is frequently wrong
  2. I have not performed any OCR on some of these documents and as such a precise/reliable OCR text layer is not present
  3. some of my PDF files are image scans of a text page and contain no OCR text layer
  4. Some of the PDFs are scans of 19th century manuscripts that have undergone some partial handwriting OCR (with variable accuracy) that I need to manually transcribe to complete.
  5. it is not possible to remove this dynamically assigned tag.
  6. the index read-through is not sufficiently thorough, probably due to items 1, 2, 3 and 4

My test:

  1. Using a small PDF that is completely unrelated to my research files and does not contain any of the Tags that I have created but in the same Finder folder as my other files.
  2. Upon importing that file into DEVONthink, I noted that no Tags were dynamically applied.
  3. I removed the file from DEVONthink and emptied the trash
  4. Then I edited the same file from Finder using PDFgear and added a few words within the PDF that would match my Tags.
  5. I Imported the new file into DEVONthink and the indexing process picks up the newly added words and dynamically applies the appropriate Tags.
  6. These Tags are identical in their behaviour as previously described in that they are not removable from the file (unless I remove the Tag from the Tag list)

How do I disable the indexing “read and tag” process so that I can apply my own tags once I know that the OCR text layer is viable? Or how do I disable the Word-List and Word-Cloud creation process?

Can I disable this suggestion process when importing files or at least choose to run it manually once I have either completed the necessary OCR or transcribing process?

These suggestions are not assigned by default. Either just ignore them or switch to another inspector if you do not need them at all.

I’m happy to ignore the Word Cloud and Word List. But how do I correct the incorrect dynamic Tag suggestions when a document is imported? A few are close but many are just way off target.

The suggestions are only that: suggestions. They’re not applied to the document.

1 Like

I understand that they are suggestions - although I don’t know why it makes the suggestions it does.

In the example below, the word “Bellairs” and “Candahar” do not appear or are even contextually inferred within the document. The word “Ships” does appear within this document but is just not contextually viable (in this instance)

I have 32 years of these particular documents and each one now has suggested dynamic Tags many of which are completely incorrect in that the Tag word either:

  1. does not appear, or
  2. is not inferred, or
  3. contextually does not apply

I currently use my manually applied Tags extensively across my documents and any incorrect items are an unnecessary/unwanted distraction.

Is it possible to do one of the following either in the current version or the next:

  1. Disable the dynamic Tags feature, or
  2. Remove the incorrect dynamic Tags

Cheers

dp

Why does it matter what DT suggests if you can just ignore it?

In the last post, for example, I have no idea which of the 10 dynamic Tags are correct or not. At the moment, I know that some of the dynamic Tags are incorrect/invalid, some of the Tags are possibly correct, while there are other Tags I wouldn’t have used if tagging manually.

No, it’s not possible to disable these suggestions nor is it possible to remove what you consider to be incorrect. These tags are suggested based on the internal AI’s assessment of the relationship the document has with other documents in your database. So they could legitimately be valid from that standpoint. However, you could make sure your local backups are current and do a File > Rebuild Database to rebuild the index from scratch and see if there’s an appreciable change. Do not interrupt the process or let the machine power down while it’s working.

perhaps it’s a problem with the dark mode. in light mode, the suggested tags have a greyed background. they are clearly distinguishable.

2 Likes

I’ll do the File>Rebuild Database as suggested.

And thanks for the definitive statement that the neither the dynamic Tag feature or its results are removable.

I’ll bite my tongue, remain silent and refrain from commenting (for a while) on the “hold-on-whose-data-is-this” concerns that this internal AI has, in my instance, arbitrarily incorrectly and inconsistently applied these permanent and unwanted Tags to my individual files compared to how I want to Tag my own individual files in my own personal database for my own use. :stuck_out_tongue: #jailbreakmydata!

I’ll report back on the results of the Database Rebuild process.

Requests and comments are certainly welcome. It’s a definite No if you don’t ask :slight_smile:

arbitrarily incorrectly and inconsistently applied these permanent and unwanted Tags to my individual files

But I will correct this again… according to everything you’ve said and shown, no tags have been applied to any of the files. If any tags had been added the Remove All button would be active.

… and in dark mode, for due diligence…

Look at your inspector again and it is clear nothing – not yourself and not AI (internal or external) – has applied tags to the selected item.

Also, you can see the applied tags on an item in the Tags bar and the Generic Info inspector / popover.
Lastly, no tags are permanent.

3 Likes

Looking at the difference between the two modes, there is a significant difference between what is suggested and what I’ve added. It also helps that your screen snippet is magnified!

Can I ask for an option to make the textual content of that inspector window larger or at least have a zoomable text factor that resembles the other windows?

And maybe an contextual option on each suggested Tag to either accept or refuse/remove the suggestion? Accepting the Tag makes it a legitimate tag that it added to both that document Tag list and the global Tag list.

:thinking:

Can we have an option to disable the AI Tagging suggestions altogether?

Please.

:innocent:

The requests are noted, with no promises.

2 Likes