Text chunk tagging

[michal-hradis]

Tagging worker

• Ask server for tagging job (list of texts to process, model to use, definition of tag - examples, name, definition, prompt)
• Process the document using OpenAI, Ollama
• Upload resulting tags to DB

Tag storage Chunks

• positive_tags list[str] - ["c1:pes", "c2:pes"]
• negative_tags list[str]
• automatic_tags list[str] (target)
• collections list[str]

DB content:
Documents -
Chunks -
Collections - user places some Chunks to a named collection, one user a owher, can share with other users

[michal-hradis]

Workflow:

• User create a tag on a collection - Name, description, examples, color, shorthand (pictogram)
• On collection run tagging task
• Tusk runs - UI shows progress (upload positive tag instances to weaviate)
• When task ends, update task state (Finished)
• User can "search" for chunks with the generated tags (within a collection)
• User can approve tag instances / disaprove tag instance
• User can bulk delete automatci tags - within a collection delete all automatic tag instances

[michal-hradis]

Task Progress

• Sql databse should hold number of chunks and number of processed chunk for each task
• UI - show table (list) of all tasks with the state (+ option to cancel)

[michal-hradis]

User Interface elements:

• Collection tag management - Functions: list with last task progress/state, create, edit (delte) tags, + run automatic tagging. + Delete all automatic + Cancel task.
• Search interface - filter by tag (positive/automatic/both), show tags for each chunk, change tags (approve, add, delete) + Delete all automatic.

[Martin-Toma]

Create tag separately from tagging task
Tagging task:

• in weaviate create new object Tag (or fetch existing one)
• llm decides if tag belongs to a chunk and store to weaviate if yes
• update task status in sql

Tag managing:

• fetch all tags - change to fetch all tags over collection
• delete selected tags - works it removes references from chunks and then removes the Tag object,
• add delete all tags in a collection

User tags: positive and negative - they are approved or disapproved automatic tags (just change reference don't store who approved it)
Run one task over a Tag at a time if user wants to run another task over the same tag, the previous task must be canceled
Don't run llm tagging over already positively approved chunk tag pairs
Rename routes
Add pydantic scheme to weaviate class Tag

link to draw.io schema

link to draw.io sequence diagrams ans schema

[michal-hradis]

• Create Tag should be a dialogue
• Existing tags should show as a list / table
• Tagging tasks should be show os separate page
• Add filter by tags directly to normal search
• Tags should be shown as badges

Create TAG badge

• Automatic
• Approved
• Deleted
• Show "X" on hover
• Click on tag approves, click on "X" deletes

[michal-hradis]

• Add UserCollection
• Show tags only within collection
• Add option to add chunks and full document to collection