Data Format

CSV Input

Chat rooms are imported via CSV. Each row represents one turn in the conversation.

turn_id,user_id,turn_text,reply_to_turn
T001,alice,"Hello everyone",
T002,bob,"Hi! How are you?",T001
T003,alice,"Doing well, thanks!",T002
T004,carol,"What are we discussing today?",

Column	Required	Description
turn_id	✅	Unique identifier for the turn within the room
user_id	✅	Speaker identifier
turn_text	✅	Turn content
reply_to_turn	❌	turn_id of the turn being replied to; leave empty if none

Tip

A sample file is provided at docs/sample_chat_room.csv.

Validation rules

• turn_id must be unique within each CSV file.
• reply_to_turn, if set, must reference a turn_id that exists in the same file.
• Empty rows and rows with missing required columns are rejected during preview.

Multi-room import

Each CSV file maps to one chat room. To import multiple rooms into a project, upload one CSV per room.

---

Export Formats

Disentanglement — JSON

{
  "project": "My Project",
  "room": "room-01",
  "annotator": "alice",
  "annotations": [
    { "turn_id": "T001", "thread_id": "A" },
    { "turn_id": "T002", "thread_id": "A" },
    { "turn_id": "T003", "thread_id": "B" }
  ]
}

One JSON file is produced per annotator per room.

Adjacency Pairs — ZIP

The export is a ZIP archive containing one plain-text file per annotator per room. Each line encodes one directed link:

T002 -> T001 [question-answer]
T004 -> T001 [topic-shift]

Format: <from_turn_id> -> <to_turn_id> [<relation_type>]