Predictive tag

Available (free) in the Microsoft Store.

Summary

An open-source app that uses a learning model to suggest descriptive metadata tags for a resource.

Features

• Uses a learning model to suggest tags that might be relevant to a current tag.
• Indicates whether suggestions are available as you type.
• Uses built-in spell checking that's configurable for your local language.
• Suggestions are cached to avoid repeated searches.
• Reminders are issued for tags on a configurable notify-list.
• Generates a comma-separated tag string from your chosen list of tags.
• Analyzes a historical dataset of previously assigned tags to generate suggestions.
• Merges team member learning models to accelerate learning.
• Exports a comma-separated synonym string.

Features under consideration

• Narrowing the suggestion search-space as users type each letter of a tag.
• JSON data format and MongoDB backend instead of text files.

Getting started

1. To get started, enter an initial tag. The highlighted Suggest button turns green when suggestions are available for that tag. Click the Suggest button or press Enter to see the suggestions.

   

2. A list of suggested tags will appear in the Suggestions box. Color-coded entries are generated by the learning model. You can search and filter suggestion results using the search box.

   

3. Double-click any suggestion in the Suggestions area to add it to your tags and promote its relevance. The learning model associates your added suggestion with the orange highlighted tag. After you've selected a suggestion, you can double-click it again to further promote it in the color-coded list.

   

   You can remove tag-suggestion associations by right-clicking a color-coded suggestion and choosing the Unrelate... option.

   

4. To move your cursor to the next tag row, press the Tab key twice.

   

5. You can also enter tags manually in the tags list. The app will learn their relevance and suggest them in the future when they are applicable.

   

   The app will add a red squiggle under misspelled tags. Right-click a misspelled tag to see one or more suggested corrections.

   

6. Optionally, select one of the numbered tag-preset buttons to add a specific set of tags. For more information, see Config.

   

7. Choose Copy to clipboard to export your tags as a comma-separated string.

App files

The app creates and uses these data files:

• historical-data.txt
• learn-data.xml
• notify-data.txt
• config.xml

Click the folder icon in the top-right corner of the app UI to open the app folder.

Historical data

The historical-data.txt file stores all previously assigned tags. Each file line contains a set of comma-separated tags (a tagset) for a single resource, such as a book or image. Future tag suggestions are drawn from the historical tags. Here's some sample tagsets for images:

Image	Tagsets	Attribution
	owl, bird, animal, fauna, species, wildlife	source
	forest, fog, trees, nature, wilderness, woods	source
	snake, reptile, animal, fauna, species, wildlife	source
	road, highway, trees, forest, asphalt, autumn, fall	source
	boats, lake, travel, vacation, outdoors, dusk, sunset	source
	beach, nature, sunset, sea, travel, vacation, dusk, outdoors	source
	girl, child, childhood, grass, flowers, meadow	source
	elephant, animal, fauna, species, wildlife	source
	sunset, sun, sea, vacation, landscape, clouds, nature	source
	oil drop, water, oil, abstract, artistic, oily	source
	sunset, fields, trees, farmland, grassland, horizon	source
	paint, painting, art, abstract	source
	paint, painting, art, abstract, acrylic paint	source
	paint, painting, art, watercolors, paintbrush, colors	source

Learn data

The learn-data.xml file stores every suggestion that a user has ever selected, along with each associated tag and the number of times the suggestion was selected for the tag. For example:

<item tag="newt" suggest="reptile" score="2" />
<item tag="owl" suggest="bird" score="3" />
<item tag="paint" suggest="art" score="6" />
<item tag="paint" suggest="paintbrush" score="5" />
<item tag="paint" suggest="watercolors" score="2" />

Notify data

The notify-data.txt file contains a list of tags that if entered by the user will generate a notification. Edit the notify-data file as needed. For example:

zonked
sloshed
gazillion

Config

The config.xml includes these configuration options:

• Set the spellcheck locale (e.g. en-US, en-GB, fr-FR):

  <language>
    <spellchecker>en-GB</spellchecker>
  </language>
  

  Your operating system must have the corresponding language pack with spellchecking installed. Language packs that aren't installed by an administrator won't include spellcheck support and your selected locale won't work. In the following Windows 11 screenshot, both English language packs have spellcheck support, but the French language pack doesn't.

  

• Configure up to five tag preset buttons:

  

  The following XML adds preset buttons 1 and 2 only. When you click preset button 1, the tag cat will appear in the tag list. Similarly, when you click preset button 2, the tags dog, puppy, and canine will appear in the tag list. Use the text attribute to set the mouse-over tip for a preset button.

  <presets>
    <preset text="Add a cat tag">
      <tag>cat</tag>
    </preset>
    <preset text="Add some dog tags">
      <tag>dog</tag>
      <tag>puppy</tag>
      <tag>canine</tag>
    </preset>
  </presets>
  

• Set the maximum number (1-250) of tags that you want to allow for a single resource:

  <tags>
    <max>75</max>
  </tags>
  

Edit the existing config.xml in the app folder, or drag and drop a new configuration file onto the Suggestions pane to replace the existing configuration file.

Accelerate learning

As you use the app, the historical-data.txt and learn-data.xml files become more comprehensive. To accelerate the app's learning process, team members can share their historical-data.txt and learn_data.xml files. To do so, simply drag and drop any data file onto the Suggestions pane. The app auto-detects the data file type (regardless of filename) and asks you whether you want to merge it into the existing same-type data file or replace the existing file.

The app automatically deduplicates during a data file merge. You don't need to restart the app after a drag & drop operation.

Error reporting

The app reports any internal errors by creating and adding a timestamped entry to an error.txt file in the app folder. Click the folder icon in the top-right corner of the app UI to open the app folder. The error.txt file isn't created unless an error occurs. The file contains debugging information, such as the source code line number where the error occured.

How tag suggestion works

1. The app filters all historical-data.txt tagsets to remove any tagset line that doesn't contain the user-entered tag (orange highlighted in the app UI).

2. The app combines all tags in the filtered tagsets into a 1D list.

3. The app sorts the list of tags in descending order of tag frequency and then removes the duplicates.

4. The app applies the following learning model to the sorted list of historical tags:

   1. When a user designates a suggestion as relevant to a tag by selecting the suggestion, the app remembers that decision and promotes that suggestion's ranking relative to other suggestions that relate to that tag.
   2. All ranked suggestions that relate to a tag are color-coded and presented in priority order when the app user requests suggestions for the tag.

5. When a user clicks Copy to clipboard on completion of tag selection for a resource, the app adds the exported tagset entry to historical-data.txt.

As a result of the app user's choices and the learning model, the historical-data and learn-data files automatically become more extensive and refined.

Manually compile a single file executable

Although it's easier to install the app from the Microsoft Store, you can compile and install the app using Visual Studio.

Prerequisites

• .NET 5+
• Visual Studio 2019+

Create an executable

1. Open the project in Visual Studio.

2. From the menu bar, choose Build > Publish...

3. Choose to publish to a folder, then select Finish.

4. Choose Show all settings.

5. Choose Framework-dependent and Produce single file in the profile settings, and note the Target location of the compiled executable.

6. Choose Publish.